Skip to content

Ears for coding agents: let AI hear, review and make sound. CPU-only, open source.

forge v2 round 1: refsets + benchmark, critic v2, search v2, dsp/fx toolkit, integrated

Leonxlnxa34fa0ed0fd2
diff --git a/OVERNIGHT.md b/OVERNIGHT.md
new file mode 100644
index 0000000..acde869
--- /dev/null
+++ b/OVERNIGHT.md
@@ -0,0 +1,38 @@
+# forge v2: overnight iteration log (branch `forge-v2`, not merged)
+
+Goal (from the user, 2026-10-16 ~02:40): iterate ~12 h to make earshot much better, above all **forge**: sound
+effects that are genuinely great for motion-design launch videos (whooshes, swishes, risers, downlifters, impacts,
+sub drops, UI clicks / taps / toggles / hovers, pops, notification dings, shimmers, glitches, typing, logo stings),
+and better outputs. At the end (~15 h): a showcase of the progress and a new clean blind test. Work only on this
+branch; the user verifies before anything reaches `main`.
+
+## How progress is measured (so the overnight work cannot fool itself)
+- A fixed **forge benchmark** (`eval/forge/`): per category, held-out real / pro reference recordings that the forge
+ never optimises against, independent text phrasings, model-free tone and envelope distances, AudioSet labels,
+ artifact checks and variant diversity. Run at the start (baseline = the v1 recipes) and after every round.
+- The forge's own critic may change; the benchmark does not (only additions, recorded here).
+- Human ears decide at the end: a clean, loudness-matched, randomised blind test.
+
+## Rounds
+| round | when | what | result |
+|---|---|---|---|
+| 0 | 02:40 | branch, plan | |
+| 1 | 02:50-05:15 | **forge v2 tooling.** Refsets: 18 motion-design categories, 81 forge / 71 bench refs, split by source family, 0 leaks (`earshot/forge/refsets.py`, `eval/forge/build_refsets.py`). Fixed benchmark `eval/forge/bench.py` + baseline. Critic v2: model-free envelope (EnvRef), artifact detectors and polish (`polish.py`), designed mode; `version=1` reproduces v1. Search v2: model-free pre-screen (tone + envelope + polish, CLAP only on the best), CMA-ES option, threaded renders, robust shortlist, seed bank, report.md / compare.png / audition.wav. DSP v2: `dsp.py` sources (air, shepard, sub_drop, bell, fm, sweeps, stack) + `fx.py` (hybrid FDN / noise-tail reverb, saturation, transient, limiter, eq_match, doppler, glitch, granular). Integration: recipes get `REFSET` (CLI default refs = its forge split), held-out guard (`refsets.held_out`: bench files and the 8 bench-source kit WAVs never reach a search), v2 components in report history / ref rows / report.md, CLI `--mode`, `--tone` default = critic v2 weight (0 = report only), critic typicality needs >= 3 refs (2 refs gave z ~1e5), compare.png envelopes smoothed for long sounds. | Tests: 236 passed, 15 skipped (full fast suite, 197 s). Smoke (critic v2 + pre-screen + forge refsets, budget 24-32): whoosh 0.7 min, impact 2.4 min; all outputs written. Baseline (v1 finals vs bench, `eval/forge/results_v1.md`): variety (CLAP cos between variants) 0.89-0.97 vs real 0.50-0.83; tail t40 x click 0.56 / whoosh 0.51 (real 0.86-1.33 / 0.60-1.08); tone dB click 7.6, pop 8.6, key 8.2, whoosh 8.6, impact 6.7 (real forge-vs-bench 11.3 / 16.1 / 8.9 / 10.3 / 6.2); model checks saturated (typ z, phrasings above real everywhere). |
+
+### Round 1 notes (integrator)
+- What works end to end: `earshot forge <recipe> [--refset CAT]` -> critic v2 inside search v2 (pre-screen reuses the
+ critic's PolishRef) -> picks re-rendered on new seeds -> WAVs + report.md + compare.png + audition.wav; then
+ `eval/forge/bench.py --made made_<tag>` against the held-out side. The smoke picks ran through the benchmark too
+ (out/forge-r1/results_smoke.md; tiny budgets, not a progress claim): whoosh var cos 0.966, t40 x 0.47 (still clones,
+ still short); impact var cos 0.72, t40 x 0.67, tone 9.1 dB (v1's 6.7 was optimistic: it used the bench impact).
+- Gaps the smoke runs show (compare.png): made whooshes reach -70 dB by ~270 ms where refs still sit at -50 dB at
+ 300-400 ms, but the critic sees no problem because the whoosh_fast *forge* refs are themselves short (real x t40 x
+ 0.6): optimising against them cannot fix the tail. Made impacts are 5-15 dB thin at 1-10 kHz and carry a narrow
+ ~70-80 Hz partial; the impact refs' tails end in the trim fade (a curved drop at 4.5-6 s), which the envelope check
+ reads as natural.
+- Known limits for round 2: impact has 2 forge refs (no typicality, LOO ref rows get no env / polish ranges, so real
+ refs score 0 penalties by construction); downlifter / glitch / sub_drop have no forge refs (designed mode, the
+ pre-screen ranks artifacts only); polish on 7 s sounds costs ~1 s each (impact pre-screen 53 s of a 2.4 min run);
+ the seed bank's impact / whoosh entries are v1 finals tuned partly against now-bench kit refs (parameters only, a
+ soft leak: start points, not refs). `eval/forge/pro/` (a pro motion-SFX study by another agent, still in progress)
+ is not part of this commit.
diff --git a/README.md b/README.md
index 32b81eb..b347a97 100644
--- a/README.md
+++ b/README.md
@@ -49,7 +49,7 @@ $ earshot critique film.mp4
| `critique FILE [--video V]` | a launch-video scorecard against a profile of 7 professional launch films, with timestamped fixes |
| `profile REFS... -o p.json` | build your own reference profile from films you admire |
| `measure FILE` | objective measurements only (no models): LUFS, true peak, LRA, envelope, spectrum, tempo / key |
-| `forge RECIPE --refs DIR` | make a sound in code and improve it by listening (see below) |
+| `forge RECIPE --refset CAT` | make a sound in code and improve it by listening (see below) |
| `llm FILE --yes-upload` | optional audio-LLM critique (Gemini), only with your own key and explicit consent |
| `mcp` | the same tools as an MCP server for Claude Code and other agents |
| `doctor [--load]` | check the install, models, data files and CPU settings |
@@ -69,23 +69,33 @@ loudness, sync to the beat):
Listening makes iteration possible. The agent writes a sound as code: a **recipe** of physical synthesis (modal
resonators, contact excitations, filtered noise, a computed room and a microphone; no recorded audio inside). The
forge renders candidates from the recipe's parameters, and a **critic** built from earshot scores each one:
-description match against named confusions, realness, closeness to real reference recordings, measured shape, and a
-model-free tonal-balance and tail check against the references. An evolutionary **search** keeps the best and mutates
-them. Then the agent reads the critic's words ("sounds like: a cork pop, a glass ting", "decay +2.4 outside the
-refs"), looks at spectrograms next to the references, rewrites the recipe, and runs it again.
+description match against named confusions, realness, closeness to real reference recordings and measured shape, then
+model-free penalties against the references: 1/3-octave tonal balance, the spectro-temporal envelope (a tail that dies
+too fast shows here), artifacts (clicks, aliasing, DC / infrasound, harsh resonances, comb ring, cut tails, clipping)
+and polish (attack, onset snap, tail smoothness). A cheap model-free **pre-screen** ranks many renders with the same
+measurements, so the models only hear the best; an evolutionary **search** keeps the best and mutates them. Then the
+agent reads the critic's words ("sounds like: a cork pop, a glass ting", "decay +2.4 outside the refs", "bumpy tail"),
+looks at spectrograms next to the references, rewrites the recipe, and runs it again.
```bash
-earshot forge # list the recipes: click, impact, keypress, pop, whoosh
-earshot forge keypress --refs path/to/real/keypresses/ --budget 144 --reps 2 --tone 1.5 --keep 4 --robust 8 --dir made/keypress
+earshot forge --list # the recipes (click, impact, keypress, pop, whoosh, ...) + refsets
+earshot forge keypress --refs path/to/real/keypresses/ --budget 144 --variants 4 --dir made/keypress
+earshot forge whoosh --refset whoosh_fast --budget 96 --dir made/whoosh # a reference set's 'forge' split
earshot forge whoosh --target "a fast crisp air whoosh" --alts "a wind gust" "a sword swing" \
- --refs path/to/real/whooshes/ --tone 1.5 --dir made/whoosh
+ --refs path/to/real/whooshes/ --mode real --dir made/whoosh
+earshot forge --render keypress --params made/keypress/report.json --seeds 1 2 3 --dir made/again
```
-Output: the best diverse sounds as WAVs plus `report.json` with every candidate's parameters and scores, the same
-critic's scores of the real references (leave-one-out), held-out AudioSet tags, and with `--robust N` each pick
-re-rendered on N new seeds. Every score also carries model-free measurements (`earshot/forge/tone.py`): the
-1/3-octave tonal balance against the references' median ("+16 dB at 500-630 Hz (above every ref)"), the tail and the
-length. In Python: `from earshot.forge import forge`.
+Output (`--dir`): the best diverse sounds as WAVs; `report.md` (what was asked, the search per generation, each pick in
+the critic's words, the measured gap to the references); `compare.png` (each pick's spectrogram, envelope and
+1/3-octave balance next to its nearest reference); `audition.wav` (made, ref, made, ref ... to listen in pairs);
+`report.json` with every candidate's parameters and scores, the same critic's scores of the real references
+(leave-one-out) and held-out AudioSet tags. Picks are re-rendered on new seeds (`--robust`, default 3) and must differ
+in parameters and in sound. Reference sets (`earshot/forge/refsets.py`) are split by source into a 'forge' side the
+search may use and a 'bench' side only the fixed benchmark (`eval/forge/bench.py`) hears. The DSP toolkit for recipes:
+`earshot/forge/dsp.py` (sources: modal, noise, air, sweeps, Shepard risers, sub drops, bells, FM; layering) and
+`earshot/forge/fx.py` (reverb, saturation, transient shaping, limiter, EQ match, Doppler, filters, glitch, granular).
+In Python: `from earshot.forge import forge`.
**Blind test.** A key press, a UI click, a pop, a fast whoosh and an impact were made from scratch in code and
played blind next to real recordings of the same kind of sound. The author picked the made one correctly **8 times out
diff --git a/earshot/forge/__init__.py b/earshot/forge/__init__.py
index 0deac4c..1e365dd 100644
--- a/earshot/forge/__init__.py
+++ b/earshot/forge/__init__.py
@@ -1,39 +1,61 @@
"""earshot forge: an agent makes its own sounds in code and improves them by listening with earshot.
- earshot forge keypress --refs kit/keyboard --budget 160 --keep 4 --dir made/keypress
- earshot forge whoosh --target "a fast crisp air whoosh" --alts "a wind gust" "a sword swing" --refs kit/whoosh
+ earshot forge keypress --refset typing_key --variants 4 --dir made/keypress # ES + pre-screen (CLI)
+ earshot forge whoosh --target "a fast crisp air whoosh" --refs kit/whoosh --strategy cma --prescreen 64
+ earshot forge --list # recipes with their TARGET (+ refset categories)
+ earshot forge --render keypress --params made/keypress/report.json --seeds 1 2 3 --dir made/again
Loop: a recipe (earshot/forge/recipes/<name>.py: physical synthesis written in code, no recorded audio) renders
-candidates from parameters; the Critic scores each one (description match z, real-or-synthetic p_real, closeness to
-real reference recordings, measured shape); an evolutionary search keeps the best and mutates them; repeat. The agent's
+candidates from parameters; a model-free pre-screen ranks many renders against the real references (tonal balance,
+envelope, artifacts: earshot.forge.search.Prescreen) and the Critic scores the best (description match z,
+real-or-synthetic p_real, closeness to real reference recordings, measured shape, `tone=` the model-free balance); a
+search (v1 evolution strategy, or separable CMA-ES with restarts) keeps the best and moves on; repeat. The agent's
outer loop reads the critic's words ("sounds like: a wooden knock", "decay +2.4 outside the refs") and rewrites the
-recipe itself. Output: the best diverse sounds as WAVs + report.json (scores, parameters, the whole history) + the same
-critic's scores of the real references (leave-one-out, so a reference is not its own neighbour) and their held-out
-AudioSet tags, so the gap to real is always visible.
+recipe itself.
+
+Output folder (`out=` / --dir): the best diverse sounds as WAVs; report.json (scores, parameters, per-generation stats,
+timing, the whole history, the same critic's scores of the real references leave-one-out and their held-out AudioSet
+tags); report.md (what was asked, budget, per-generation best, the critic's words for each pick, the measured gap to
+the references); compare.png (each pick's spectrogram, envelope and 1/3-octave balance next to its nearest reference:
+earshot.forge.viz); audition.wav + audition.txt (made, ref, made, ref ... peak-matched, to listen in pairs).
+The seed bank (earshot/forge/seeds/<recipe>.json) keeps good parameter sets; `bank=True` (CLI default) starts the
+search from them and adds the new picks. `shortlist=S` with `robust=N` re-renders S x keep diverse candidates on N new
+seeds and keeps the `keep` best robust means (the best single render of a search is a winner's curse).
+
+Python API defaults reproduce v1 (strategy 'es', no pre-screen, no bank), so earlier round scripts stay comparable;
+the CLI defaults to the v2 setup (es + prescreen 3 x pop, bank, robust 3, shortlist 2, critic v2 with its own weights,
+and when neither --refs nor --refset is given the recipe's REFSET, else its REFS minus held-out files). The critic is
+v2 (earshot/forge/critic.py): the v1 model terms minus model-free tone, envelope, artifact and polish penalties; the
+pre-screen ranks with the same tone / polish measurements, so CLAP only hears the best `pop` of each generation. Measured on keypress at 96 scored candidates: the pre-screen raised the
+passing candidates from 10.5 to 36.5 and the top-5 mean total from 5.77 to 6.17; CMA-ES did not beat the v1 ES at
+96-240 candidates (earshot/forge/search.py has the numbers).
Honest limits: the critic's models can be pleased without the sound getting better (Goodhart). earshot's realness model
is easy to fool with modal resonators, a room and a mic noise floor, so a high p_real proves little. Worse, once a
search has tuned against them, every model check saturates: the forge test's made sounds land inside or above the real
recordings on held-out phrasings, the AudioSet tagger and independent real sets too, while a plain 1/3-octave
measurement still finds them 8-13 dB off the real balance. So the critic always reports the model-free tone check
-(earshot/forge/tone.py: tonal balance against the refs, the bands outside every ref, the tail), and `tone=W` (CLI
---tone W, 1.5 is a good start) puts it into the objective. Judge finished sounds by those measurements, robust
-re-renders (`robust=N`), spectrograms and, last, human ears. AudioSet tags are a held-out check only while nobody picks
-or steers by them: once they decide which candidate is kept, they are part of the objective.
+(earshot/forge/tone.py), the pre-screen ranks by it, and critic v2 always puts it into the objective (weight 1.0;
+`tone=W` / --tone W changes it, --tone 0 makes it report-only). Judge finished sounds by those measurements, robust
+re-renders, compare.png, the held-out benchmark (eval/forge/bench.py) and, last, human ears (audition.wav). AudioSet
+tags are a held-out check only while nobody picks or steers by them.
"""
import importlib
import json
import os
import pkgutil
+import time
import numpy as np
from .. import audio
from ..report import emit
+from . import recipes as _recipes_pkg # first import binds earshot.forge.recipes to the package; the function below
+ # rebinds it once, for good (later imports hit sys.modules and leave it)
def recipes():
- pkg = importlib.import_module(f'{__name__}.recipes') # not `from . import recipes`: this function shadows it
+ pkg = _recipes_pkg
return sorted(m.name for m in pkgutil.iter_modules(pkg.__path__) if not m.name.startswith('_'))
@@ -41,57 +63,274 @@ def load_recipe(name):
return importlib.import_module(f'earshot.forge.recipes.{name}')
+def pick_refs(refs=None, refset=None, recipe=None, log=None):
+ """(reference files, refset category used) for a run: `refset` (a category of earshot.forge.refsets, its 'forge'
+ split only: the 'bench' split stays held out), else `refs` (files / folders), else the recipe's REFSET when this
+ machine has a registry with refs in it, else the recipe's REFS that exist here. Held-out files (a registry bench
+ ref anywhere, or for the recipe's REFS a kit file cut from a bench source: refsets.held_out) are dropped."""
+ from . import refsets as _rs
+
+ def keep(paths, kit):
+ bad = set(_rs.held_out(paths, kit=kit))
+ if bad and log:
+ log(f"dropped {len(bad)} held-out reference(s) (benchmark side): "
+ + ', '.join(sorted(os.path.basename(p) for p in bad)))
+ return [p for p in paths if p not in bad]
+
+ if refset:
+ return _rs.refset(refset, split='forge'), refset
+ if refs:
+ return keep(audio.list_media(refs, exts=audio.AUDIO_EXTS), kit=False) or None, None
+ cat = getattr(recipe, 'REFSET', None)
+ if cat:
+ try:
+ rr = _rs.refset(cat, split='forge')
+ if rr:
+ return rr, cat
+ except (_rs.RefsetsMissing, KeyError):
+ pass
+ rr = keep([p for p in (getattr(recipe, 'REFS', None) or []) if os.path.exists(p)], kit=True)
+ return rr or None, None
+
+
+def resolve_refs(refs=None, refset=None, recipe=None, log=None):
+ """the reference files of pick_refs() (None when there are none)"""
+ return pick_refs(refs, refset, recipe, log)[0]
+
+
+# score keys copied into report['history'] and report['ref_scores'] (v1 model terms + the v2 model-free components)
+SCORE_KEYS = ('total', 'passes', 'match_z', 'alt_margin', 'p_real', 'ref_typ_z', 'feat_pen', 'tone_db', 'tone_ratio',
+ 'sounds_like', 'total_v1', 'mode', 'tone_pen', 'env_db', 'env_ratio', 'env_out', 'env_pen', 'art_pen',
+ 'polish_pen', 'hard', 'polish_words')
+
+
+def _fmt(v, nd=2):
+ return '-' if v is None else (f'{v:.{nd}f}' if isinstance(v, (float, np.floating)) else str(v))
+
+
+def audition(path, made, refs, pairs, gap_s=0.6, pair_gap_s=1.2, peak_db=-3.0):
+ """write made, ref, made, ref ... (each peak-normalised to peak_db, stereo 48 kHz) + a .txt cue list next to it;
+ made / refs = [(name, array)], pairs = a ref index (or None) per made. Returns {'file', 'txt', 'cues'}"""
+ import soundfile as sf
+ from . import dsp
+
+ def st(x):
+ x = np.asarray(x, np.float64)
+ x = np.stack([x, x], 1) if x.ndim == 1 else x
+ return dsp.normalize(x, peak_db)
+
+ parts, cues, t = [], [], 0.0
+ for i, (name, x) in enumerate(made):
+ seq = [('made', name, x)]
+ j = pairs[i] if i < len(pairs) else None
+ if j is not None:
+ seq.append(('ref', refs[j][0], refs[j][1]))
+ for k, (kind, nm, y) in enumerate(seq):
+ y = st(y)
+ cues.append({'t': round(t, 2), 'kind': kind, 'name': nm})
+ parts.append(y)
+ t += len(y) / 48000
+ g = gap_s if k < len(seq) - 1 else pair_gap_s
+ parts.append(np.zeros((int(g * 48000), 2)))
+ t += g
+ x = np.concatenate(parts) if parts else np.zeros((4800, 2))
+ sf.write(path, x.astype(np.float32), 48000, subtype='PCM_24')
+ txt = os.path.splitext(path)[0] + '.txt'
+ lines = [f'{os.path.basename(path)}: each made sound, then its nearest real reference (measured: tonal balance + '
+ f'envelope), every sound peak-normalised to {peak_db:g} dBFS (peak-matched, not loudness-matched)']
+ lines += [f"{c['t']:8.2f} s {c['kind']:<4} {c['name']}" for c in cues]
+ with open(txt, 'w', encoding='utf-8') as fh:
+ fh.write('\n'.join(lines) + '\n')
+ return {'file': path.replace(chr(92), '/'), 'txt': txt.replace(chr(92), '/'), 'cues': cues}
+
+
+def report_md(r):
+ """the run as Markdown: what was asked, the budget, per-generation best, each pick in the critic's words and the
+ measured gap to the real references"""
+ L = [f"# forge {r['recipe']}: '{r['target']}'", '']
+ refs = [os.path.basename(p) for p in r.get('refs') or []]
+ tm = r.get('timing') or {}
+ L.append(f"**Asked:** recipe `{r['recipe']}`, target '{r['target']}'"
+ + (f", alternatives: {', '.join(repr(a) for a in r['alts'])}" if r.get('alts') else '')
+ + (f", refset `{r['refset']}`" if r.get('refset') else '')
+ + f"; {len(refs)} reference recording(s)" + (f": {', '.join(refs)}" if refs else '')
+ + (f"; critic v{r['critic']['version']}, mode {r['critic']['mode']}" if r.get('critic') else '') + '.')
+ L.append('')
+ L.append(f"**Budget:** strategy `{r.get('strategy', 'es')}`, {r['budget']} critic-scored candidates "
+ f"(pop {r.get('pop')}, {r['generations']} generations)"
+ + (f", pre-screen: {r['prescreen']} renders per generation ranked by measurement, the best "
+ f"{r.get('pop')} scored" if r.get('prescreen') else ', no pre-screen')
+ + f", reps {r.get('reps', 1)}, tone weight {r.get('tone_weight')}, seed {r['seed']}"
+ + (f", {r.get('n_bank', 0)} start(s) from the seed bank" if r.get('n_bank') else '')
+ + (f", {r.get('restarts')} restart(s)" if r.get('restarts') else '') + '.')
+ if tm:
+ L.append(f"Took {tm.get('total_s', 0) / 60:.1f} min for the search: {tm.get('rendered')} renders, "
+ f"{tm.get('scored')} scored ({tm.get('scored_per_min')} scored / min, {tm.get('rendered_per_min')} "
+ f"rendered / min; render {tm.get('render_s')} s, measurements {tm.get('pre_s')} s, critic "
+ f"{tm.get('score_s')} s).")
+ L += ['', '## Search', '', '| gen | best total | this gen best | this gen median | passes | best tone dB | sigma | note |',
+ '|---|---|---|---|---|---|---|---|']
+ for g in r.get('gens', []):
+ L.append(f"| {g['gen']} | {_fmt(g.get('best'))} | {_fmt(g.get('gen_best'))} | {_fmt(g.get('gen_median'))} | "
+ f"{g.get('passes')}/{g.get('scored')} | {_fmt(g.get('best_tone_db'), 1)} | {_fmt(g.get('sigma'), 3)} | "
+ f"{('restart from ' + g['restart']) if g.get('restart') else ''} |")
+ L += ['', '## Picks', '', r.get('pick_rule', ''), '']
+ for p in r['picks']:
+ L.append(f"### {p['name']}" + (f" ({os.path.basename(p['file'])})" if p.get('file') else ''))
+ L.append(f"- critic: {p['score'].get('summary')}")
+ rb = p.get('robust')
+ if rb:
+ L.append(f"- robust ({rb['n']} new seeds): mean {_fmt(rb['mean'])}, min {_fmt(rb['min'])}, sd {_fmt(rb['sd'])}, "
+ f"pass rate {_fmt(rb['pass_rate'])}" + (f", tone {rb['tone_db_mean']} dB" if 'tone_db_mean' in rb else ''))
+ m = p.get('measured') or {}
+ if m:
+ L.append(f"- measured vs the refs: tone {_fmt(m.get('tone_db'), 1)} dB from their median (a typical real one "
+ f"sits {_fmt(m.get('ref_spread'), 1)} dB away), t40 {_fmt(m.get('t40_ms'), 0)} ms (refs "
+ f"{m.get('t40_ref')}), length {_fmt(m.get('length_s'))} s (refs {m.get('length_ref')}), nearest ref by "
+ f"measurement: {m.get('nearest_ref')}, by CLAP: {p['score'].get('nearest_ref')}"
+ + (f"; artifacts: {m['art']}" if m.get('art') else ''))
+ if p.get('holdout'):
+ L.append(f"- AudioSet (held out, decided nothing): {p['holdout'].get('audioset_top')}")
+ L.append('- params: `' + json.dumps({k: (round(v, 5) if isinstance(v, float) else v)
+ for k, v in p['params'].items()}) + f"` seed {p['seed']}")
+ L.append('')
+ if r.get('ref_scores'):
+ L += ['## Gap to the real references', '',
+ 'The same critic scores the references leave-one-out (a reference is not its own neighbour). Model '
+ 'checks saturate once a search has tuned against them: read the measured columns first.', '',
+ '| sound | total | passes | match z | p_real | tone dB | env pen | art pen | polish pen | hard | t40 ms | '
+ 'length s |', '|---|---|---|---|---|---|---|---|---|---|---|---|']
+
+ def v2(s):
+ return (f"{_fmt(s.get('env_pen'))} | {_fmt(s.get('art_pen'))} | {_fmt(s.get('polish_pen'))} | "
+ f"{', '.join(s.get('hard') or []) or '-'}")
+ for p in r['picks']:
+ s, m = p['score'], p.get('measured') or {}
+ L.append(f"| made {p['name']} | {_fmt(s.get('total'))} | {s.get('passes')} | {_fmt(s.get('match_z'))} | "
+ f"{_fmt(s.get('p_real'))} | {_fmt(m.get('tone_db'), 1)} | {v2(s)} | {_fmt(m.get('t40_ms'), 0)} | "
+ f"{_fmt(m.get('length_s'))} |")
+ for s in r['ref_scores']:
+ L.append(f"| real {s['file']} | {_fmt(s.get('total'))} | {s.get('passes')} | {_fmt(s.get('match_z'))} | "
+ f"{_fmt(s.get('p_real'))} | {_fmt(s.get('tone_loo_db'), 1)} | {v2(s)} | {_fmt(s.get('t40_ms'), 0)} | "
+ f"{_fmt(s.get('length_s'))} |")
+ if r.get('gap'):
+ L += ['', r['gap']]
+ L += ['', '## Files', '', '- WAVs: ' + ', '.join(os.path.basename(p['file']) for p in r['picks'] if p.get('file')),
+ '- compare.png: each pick next to its nearest reference (spectrograms, envelopes, 1/3-octave balance)',
+ '- audition.wav + audition.txt: made, ref, made, ref ... peak-matched',
+ '- report.json: everything above plus the whole search history', '']
+ return '\n'.join(L)
+
+
def forge(recipe, target=None, alts=None, refs=None, budget=160, pop=16, keep=4, seed=0, out=None, log=print,
init=None, holdout=True, robust=0, space=None, sigma0=0.22, sigma1=0.04, critic=None, tone=0.0,
- tone_target=None, reps=1):
- """run the loop for one recipe; writes WAVs + report.json into `out` when given; returns the report dict.
- robust=N re-renders every pick on N new seeds (mean / min / pass rate in pick['robust']); space= a narrower
- {name: spec} family; sigma0/sigma1 the first / last mutation step; critic= a prebuilt Critic (reuse one across
- several forge() calls in a round script, or pass one with an `extra` check); tone= the weight of the model-free
- tonal-balance penalty (Critic tone=, tone_target=); reps= renders per candidate inside the search (ranked by
- their mean, against lucky single renders)"""
+ tone_target=None, reps=1, strategy='es', prescreen=0, workers=None, bank=False, shortlist=1, refset=None,
+ outputs=True, bank_dir=None, weights=None, mode='auto'):
+ """run the loop for one recipe; writes WAVs + report.json (+ report.md, compare.png, audition.wav when outputs)
+ into `out` when given; returns the report dict.
+ robust=N re-renders every pick on N new seeds (mean / min / pass rate in pick['robust']); shortlist=S (with
+ robust) robust-scores S x keep diverse candidates and keeps the keep best robust means; space= a narrower
+ {name: spec} family; sigma0/sigma1 the first / last ES mutation step; critic= a
+ prebuilt Critic; tone= the weight of the model-free tonal-balance penalty (Critic tone=, tone_target=); reps=
+ renders per candidate inside the search; strategy 'es' | 'cma' (sigma0 / sigma1 are the ES steps; CMA starts at
+ 0.15); prescreen=K renders per generation ranked by
+ measurement before the critic (K > pop); workers= render threads; bank=True starts from the recipe's seed bank and
+ adds the picks to it (bank_dir= another folder); refset= a refsets category (its 'forge' split) instead of refs
+ (held-out bench files among explicit refs are dropped: pick_refs); weights= / mode= go to the Critic (v2 weights,
+ 'auto' | 'real' | 'designed'; weights={'tone': 0} makes the tone check report-only)"""
import soundfile as sf
from .critic import Critic
- from .search import diverse, optimize, render
- from .search import robust as robust_stats
+ from .search import diverse, optimize, prescreen_for, render, robust as robust_stats, sound_dist
+ t0 = time.time()
rec = load_recipe(recipe) if isinstance(recipe, str) else recipe
name = rec.__name__.rsplit('.', 1)[-1]
if critic is None:
target = target or rec.TARGET
alts = list(alts if alts is not None else getattr(rec, 'ALTS', []))
- critic = Critic(target, alts, refs, tone=tone, tone_target=tone_target)
+ refs = resolve_refs(refs, refset, log=log) if (refs or refset) else refs
+ critic = Critic(target, alts, refs, tone=tone, tone_target=tone_target, weights=weights, mode=mode)
else:
target, alts = critic.target, list(critic.alts)
- res = optimize(rec, critic, budget=budget, pop=pop, seed=seed, init=init, log=log, space=space, sigma0=sigma0,
- sigma1=sigma1, reps=reps)
- picks = diverse(res['best'] + sorted(res['history'], key=lambda h: -h['score']['total'])[:64], rec, keep)
+ real = name in recipes()
+ starts = list(init or [])
+ n_bank = 0
+ if bank and real:
+ from .search import load_seeds
+ bs = load_seeds(name, target, critic.refs, folder=bank_dir)
+ n_bank = len(bs)
+ starts += bs
+ pre = prescreen_for(critic)
+ res = optimize(rec, critic, budget=budget, pop=pop, seed=seed, init=starts, log=log, space=space, sigma0=sigma0,
+ sigma1=sigma1, reps=reps, strategy=strategy, prescreen=prescreen, workers=workers, pre=pre)
+ pool = res['best'] + sorted(res['history'], key=lambda h: -h['score']['total'])[:64]
+ S = max(1, int(shortlist)) if robust else 1
+ picks = diverse(pool, rec, keep * S)
+ rstats = [robust_stats(rec, critic, h['params'], robust, workers=workers) for h in picks] if robust else []
+ if robust and S > 1:
+ order = sorted(range(len(picks)), key=lambda i: (-(rstats[i]['pass_rate'] >= 0.5),
+ -(rstats[i]['mean'] if rstats[i]['mean'] is not None else -1e9)))
+ order = order[:keep]
+ picks, rstats = [picks[i] for i in order], [rstats[i] for i in order]
+ rule = (f'Chosen by a fixed rule: {keep * S} diverse strong candidates (parameters and sound differ), re-rendered '
+ f'on {robust} new seeds each, the {keep} with the best robust mean kept (passing at least half first).')
+ else:
+ rule = ('Chosen by a fixed rule: the strongest candidates whose parameters and sound differ (passing first)'
+ + (f', each re-rendered on {robust} new seeds for the robust columns.' if robust else '.'))
ref_scores = critic.score_refs(loo=True) if critic.refs else []
- report = {'recipe': name, 'target': target, 'alts': alts, 'refs': critic.refs, 'budget': budget, 'seed': seed,
- 'tone_weight': getattr(critic, 'tone', 0.0), 'tone_target': getattr(critic, 'tone_target', None),
- 'reps': reps,
- 'generations': res['generations'], 'picks': [], 'ref_scores': [], 'ref_scores_loo': True}
+ report = {'recipe': name, 'target': target, 'alts': alts, 'refs': critic.refs, 'refset': refset, 'budget': budget,
+ 'seed': seed, 'pop': pop, 'tone_weight': (critic.W.get('tone') if getattr(critic, 'version', 1) >= 2 and hasattr(critic, 'W')
+ else getattr(critic, 'tone', 0.0)),
+ 'tone_target': getattr(critic, 'tone_target', None), 'reps': reps, 'strategy': strategy,
+ 'critic': {'version': getattr(critic, 'version', 1), 'mode': getattr(critic, 'mode', None),
+ 'weights': getattr(critic, 'W', None)},
+ 'prescreen': res['prescreen'], 'restarts': res.get('restarts', 0), 'n_bank': n_bank,
+ 'generations': res['generations'], 'gens': res['gens'], 'timing': res['timing'], 'pick_rule': rule,
+ 'picks': [], 'ref_scores': [], 'ref_scores_loo': True}
if space is not None:
report['space'] = space
xs = [render(rec, h['params'], h['seed']) for h in picks]
hold = critic.holdout([x for x in xs if x is not None]) if holdout and xs else []
- ref_hold = critic.holdout([audio.load(p) for p in critic.refs]) if holdout and critic.refs else []
+ ref_x = [audio.load(p) for p in critic.refs] if critic.refs else []
+ ref_hold = critic.holdout(ref_x) if holdout and ref_x else []
if out:
os.makedirs(out, exist_ok=True)
+ pairs = []
for i, (h, x) in enumerate(zip(picks, xs)):
item = {'name': f'{name}_{i + 1:02d}', 'params': h['params'], 'seed': h['seed'],
'score': {k: v for k, v in h['score'].items() if k != 'i'}}
if i < len(hold):
item['holdout'] = hold[i]
if robust:
- item['robust'] = robust_stats(rec, critic, h['params'], robust)
+ item['robust'] = rstats[i]
+ j = None
+ if x is not None and pre.n:
+ row = pre.row(x)
+ j = pre.nearest(row['desc'])
+ d = row['desc']
+ item['measured'] = {'tone_db': row['tone_db'], 'tone_ratio': row['tone_ratio'],
+ 'ref_spread': round(pre.spread, 2), 'env_z': row['env_z'],
+ 't40_ms': d['env']['t40_ms'], 'length_s': round(len(x) / 48000, 3),
+ 't40_ref': [round(min(r['env']['t40_ms'] for r in pre.descs)),
+ round(max(r['env']['t40_ms'] for r in pre.descs))],
+ 'length_ref': [round(min(len(r) for r in ref_x) / 48000, 3),
+ round(max(len(r) for r in ref_x) / 48000, 3)],
+ 'nearest_ref': os.path.basename(critic.refs[j]),
+ 'nearest_dist': round(sound_dist(d, pre.descs[j]), 2), 'art': row['art']['flags']}
+ pairs.append(j)
if out and x is not None:
item['file'] = os.path.join(out, item['name'] + '.wav').replace(chr(92), '/')
sf.write(item['file'], x, 48000, subtype='PCM_24')
report['picks'].append(item)
for i, (p, s) in enumerate(zip(critic.refs, ref_scores)):
- report['ref_scores'].append({'file': os.path.basename(p), **{k: s.get(k) for k in
- ('total', 'passes', 'match_z', 'alt_margin', 'alt_zs', 'p_real', 'ref_typ_z',
- 'feat_pen', 'tone_db', 'tone_ratio', 'sounds_like')}})
+ report['ref_scores'].append({'file': os.path.basename(p), 'alt_zs': s.get('alt_zs'),
+ **{k: s.get(k) for k in SCORE_KEYS}})
+ if pre.n > i:
+ others = [pre.descs[k]['profile'] for k in range(pre.n) if k != i]
+ if others:
+ report['ref_scores'][-1]['tone_loo_db'] = round(float(np.sqrt(np.mean(
+ (pre.descs[i]['profile'] - np.median(others, 0)) ** 2))), 2)
+ report['ref_scores'][-1]['t40_ms'] = pre.descs[i]['env']['t40_ms']
+ report['ref_scores'][-1]['length_s'] = round(len(ref_x[i]) / 48000, 3)
if i < len(ref_hold):
report['ref_scores'][-1]['holdout'] = ref_hold[i]
best_ref = max((r['total'] for r in report['ref_scores'] if r['total'] is not None and np.isfinite(r['total'])),
@@ -102,44 +341,167 @@ def forge(recipe, target=None, alts=None, refs=None, budget=160, pop=16, keep=4,
f"(best total {top.get('total')}, match z {top.get('match_z')}, p_real {top.get('p_real')})"
+ (f"; the real references score up to {best_ref:.2f} with the same critic" if best_ref is not None
else '') + ' (references scored leave-one-out).')
+ md = [p['measured']['tone_db'] for p in report['picks'] if p.get('measured')]
+ rl = [r['tone_loo_db'] for r in report['ref_scores'] if r.get('tone_loo_db') is not None]
+ if md and rl:
+ report['gap'] = (f"Measured gap: the picks sit {min(md):.1f}-{max(md):.1f} dB from the references' median "
+ f"1/3-octave balance; the references themselves sit {min(rl):.1f}-{max(rl):.1f} dB from the "
+ f"others' median (leave-one-out). Inside that range = as typical as a real recording.")
+ report['summary'] += ' ' + report['gap']
report['history'] = [{'gen': h['gen'], 'params': h['params'], 'seed': h['seed'],
- 'score': {k: h['score'].get(k) for k in ('total', 'passes', 'match_z', 'alt_margin', 'p_real',
- 'ref_typ_z', 'feat_pen', 'tone_db', 'tone_ratio',
- 'sounds_like')}}
+ 'score': {k: h['score'].get(k) for k in SCORE_KEYS},
+ **({'pre': h['pre']} if 'pre' in h else {})}
for h in res['history']]
+ report['timing']['forge_s'] = round(time.time() - t0, 1)
if out:
+ report['files'] = {}
+ if outputs:
+ from .viz import compare_png
+ made = [(p['name'], x) for p, x in zip(report['picks'], xs) if x is not None]
+ refn = [(os.path.basename(p), x) for p, x in zip(critic.refs, ref_x)]
+ pr = [j for j, x in zip(pairs, xs) if x is not None]
+ try:
+ report['files']['compare'] = compare_png(os.path.join(out, 'compare.png'), made, refn, pairs=pr,
+ title=f"forge {name}: made (left) vs nearest real reference")['file']
+ except Exception as e: # a plotting problem must not lose the run
+ report['files']['compare_error'] = f'{type(e).__name__}: {e}'
+ report['files']['audition'] = audition(os.path.join(out, 'audition.wav'), made, refn, pr)['file']
+ with open(os.path.join(out, 'report.md'), 'w', encoding='utf-8') as fh:
+ fh.write(report_md(report))
+ report['files']['report_md'] = os.path.join(out, 'report.md').replace(chr(92), '/')
with open(os.path.join(out, 'report.json'), 'w', encoding='utf-8') as fh:
json.dump(report, fh, indent=1, default=lambda o: o.tolist() if hasattr(o, 'tolist') else str(o))
+ if bank and real and report['picks']:
+ from .search import save_seeds
+ try:
+ save_seeds(name, report['picks'], target, critic.refs, folder=bank_dir,
+ source=f"forge {strategy} budget {budget} seed {seed}" + (f' refset {refset}' if refset else ''))
+ except OSError as e:
+ if log:
+ log(f'seed bank not updated: {e}')
return report
+def render_params(recipe, params, seeds=None, out='.', refs=None):
+ """render parameter sets without searching. params: a dict, a list of dicts, a report.json dict (its picks, with
+ their own seeds unless `seeds` is given) or a seed bank dict (its entries). Writes <out>/<recipe>_r<i>_s<seed>.wav;
+ with refs, each file also gets the model-free measurements (tone dB, t40, nearest ref). Returns {'recipe', 'files'}"""
+ import soundfile as sf
+ from .search import Prescreen, render
+ rec = load_recipe(recipe) if isinstance(recipe, str) else recipe
+ name = rec.__name__.rsplit('.', 1)[-1]
+ if isinstance(params, dict) and 'picks' in params:
+ sets = [(p['params'], p.get('seed')) for p in params['picks']]
+ elif isinstance(params, dict) and 'entries' in params:
+ sets = [(e['params'], None) for e in params['entries']]
+ elif isinstance(params, dict):
+ sets = [(params, None)]
+ else:
+ sets = [(p, None) for p in params]
+ ref_files = audio.list_media(refs, exts=audio.AUDIO_EXTS) if refs else []
+ pre = Prescreen(ref_files) if ref_files else None
+ os.makedirs(out, exist_ok=True)
+ files = []
+ for i, (p, own) in enumerate(sets):
+ for s in (seeds or [own if own is not None else 0]):
+ x = render(rec, {**(getattr(rec, 'DEFAULTS', {}) or {}), **p}, int(s))
+ item = {'set': i, 'seed': int(s), 'file': None}
+ if x is not None:
+ item['file'] = os.path.join(out, f'{name}_r{i + 1:02d}_s{int(s)}.wav').replace(chr(92), '/')
+ sf.write(item['file'], x, 48000, subtype='PCM_24')
+ if pre is not None and pre.n:
+ row = pre.row(x)
+ item['measured'] = {'tone_db': row['tone_db'], 'tone_ratio': row['tone_ratio'],
+ 't40_ms': row['desc']['env']['t40_ms'], 'art': row['art']['flags'],
+ 'nearest_ref': os.path.basename(ref_files[pre.nearest(row['desc'])])}
+ files.append(item)
+ return {'recipe': name, 'files': files}
+
+
def add_cli(sub):
from ..cli import common
p = common(sub.add_parser('forge', help='make sounds in code and improve them by listening (recipe + critic + search)'))
p.add_argument('recipe', nargs='?', help='recipe name (earshot/forge/recipes/); omit to list them')
+ p.add_argument('--list', action='store_true', help='list the recipes with their TARGET (and refset categories)')
+ p.add_argument('--render', nargs='?', const=True, default=None, metavar='RECIPE',
+ help='render --params without searching (the recipe may follow --render or be the positional)')
+ p.add_argument('--params', help='JSON: a params dict, a list of them, a report.json or a seed bank file')
+ p.add_argument('--seeds', nargs='*', type=int, help='noise seeds for --render (default: each pick\'s own, else 0)')
p.add_argument('--target', help="description to aim for (default: the recipe's)")
p.add_argument('--alts', nargs='*', help='confusions it must sound less like')
p.add_argument('--refs', nargs='*', help='real reference recordings (files / folders) to compare with')
- p.add_argument('--budget', type=int, default=160, help='candidates to render and score (default 160)')
+ p.add_argument('--refset', help="a reference-set category (earshot.forge.refsets; its 'forge' split)")
+ p.add_argument('--budget', type=int, default=160, help='candidates to score with the critic (default 160)')
p.add_argument('--pop', type=int, default=16)
- p.add_argument('--keep', type=int, default=4, help='diverse best sounds to write')
+ p.add_argument('--variants', '--keep', dest='keep', type=int, default=4, help='diverse best sounds to write')
p.add_argument('--seed', type=int, default=0)
- p.add_argument('--dir', dest='outdir', help='folder for the WAVs + report.json')
- p.add_argument('--robust', type=int, default=0, help='re-render every kept sound on N new seeds and report the mean')
- p.add_argument('--tone', type=float, default=0.0,
- help='weight of the model-free tonal-balance penalty vs the refs (0 = report only; 1.5 is a good start)')
+ p.add_argument('--dir', dest='outdir', help='folder for the WAVs, report.json / .md, compare.png, audition.wav')
+ p.add_argument('--strategy', choices=('es', 'cma'), default='es',
+ help='search: es (v1 evolution strategy, default: it beat CMA-ES at these budgets) or cma')
+ p.add_argument('--prescreen', type=int, default=None,
+ help='renders per generation ranked by measurement before the critic (default 3 x pop; 0 = off)')
+ p.add_argument('--robust', type=int, default=3, help='re-render every kept sound on N new seeds (default 3; 0 = off)')
+ p.add_argument('--shortlist', type=int, default=2,
+ help='with --robust: robust-score this many x --variants candidates, keep the best robust means')
+ p.add_argument('--no-bank', action='store_true', help='do not start from / add to the seed bank')
+ p.add_argument('--workers', type=int, default=None, help='render threads (default EARSHOT_THREADS or 2)')
+ p.add_argument('--tone', type=float, default=None,
+ help="weight of the model-free tonal-balance penalty vs the refs (default: the critic's, 1.0 in v2; "
+ "0 = report only)")
+ p.add_argument('--mode', choices=('auto', 'real', 'designed'), default='auto',
+ help='critic mode: real refs (realness counts) or designed SFX (realness reported, not required)')
p.add_argument('--reps', type=int, default=1,
help='renders per candidate inside the search, ranked by their mean (2 = half as lucky, twice as slow)')
p.add_argument('--tone-target', type=float, default=None,
help="dB distance from the refs' median balance that counts as typical (default: the refs' own spread)")
def run(a):
- if not a.recipe:
- print('recipes: ' + ', '.join(recipes()))
+ if a.render is not None:
+ name = a.render if isinstance(a.render, str) else a.recipe
+ if not name or not a.params:
+ print('usage: earshot forge --render <recipe> --params file.json [--seeds 1 2 3] [--dir out]')
+ return 2
+ with open(a.params, encoding='utf-8') as fh:
+ params = json.load(fh)
+ refs = resolve_refs(a.refs, a.refset, log=print)
+ r = render_params(name, params, a.seeds, a.outdir or '.', refs)
+ lines = [f"rendered {sum(1 for f in r['files'] if f['file'])} sound(s) of {r['recipe']}:"]
+ for f in r['files']:
+ m = f.get('measured')
+ lines.append(f" {f['file'] or 'failed or silent'}" + (
+ f" tone {m['tone_db']} dB, t40 {m['t40_ms']:.0f} ms, nearest {m['nearest_ref']}" if m else ''))
+ emit(r, a, text='\n'.join(lines))
+ return 0
+ if a.list or not a.recipe:
+ lines = ['recipes (name: TARGET):']
+ for n in recipes():
+ try:
+ m = load_recipe(n)
+ lines.append(f' {n}: {m.TARGET}' + (f' [refset {m.REFSET}]' if getattr(m, 'REFSET', None) else ''))
+ except Exception as e:
+ lines.append(f' {n}: (does not load: {type(e).__name__})')
+ try:
+ from .refsets import categories
+ lines.append('refsets: ' + ', '.join(categories()))
+ except Exception:
+ lines.append('refsets: none on this machine (see earshot/forge/refsets.py)')
+ emit({'recipes': recipes()}, a, text='\n'.join(lines))
return 0
- r = forge(a.recipe, a.target, a.alts, a.refs, a.budget, a.pop, a.keep, a.seed, a.outdir, robust=a.robust,
- tone=a.tone, tone_target=a.tone_target, reps=a.reps)
+ rec = load_recipe(a.recipe)
+ refs, cat = pick_refs(a.refs, a.refset, rec, log=print)
+ if not refs:
+ print('note: no reference recordings (--refs / --refset): the critic cannot compare with real sounds')
+ elif cat and not a.refset:
+ print(f"refs: the recipe's refset '{cat}' ({len(refs)} forge-split recordings)")
+ pre = a.prescreen if a.prescreen is not None else 3 * a.pop
+ tone, weights = (a.tone or 0.0), ({'tone': 0.0} if a.tone == 0 else None)
+ r = forge(rec, a.target, a.alts, refs, a.budget, a.pop, a.keep, a.seed, a.outdir, robust=a.robust,
+ tone=tone, tone_target=a.tone_target, reps=a.reps, strategy=a.strategy, prescreen=pre,
+ workers=a.workers, bank=not a.no_bank, shortlist=a.shortlist, refset=cat, weights=weights,
+ mode=a.mode)
lines = [r['summary']] + [f" {p['name']}: {p['score'].get('summary')}" for p in r['picks']]
+ if a.outdir:
+ lines.append(f"wrote {a.outdir}: WAVs, report.json, report.md, compare.png, audition.wav")
emit({k: v for k, v in r.items() if k != 'history'}, a, text='\n'.join(lines))
return 0
p.set_defaults(func=run)
diff --git a/earshot/forge/critic.py b/earshot/forge/critic.py
index e24d4e5..6350c46 100644
--- a/earshot/forge/critic.py
+++ b/earshot/forge/critic.py
@@ -5,51 +5,79 @@
refs='refs/keyboard/')
rows = c.score([x1, x2]) # 48 kHz arrays (mono or stereo)
rows[0]['total'], rows[0]['passes'], rows[0]['summary']
+ Critic('a glitchy digital stutter', mode='designed') # no refs / designed refs: realness not required
+ c.consistency([x1, x2, x3]) # do the variants play at one loudness?
One CLAP embedding per sound feeds every model judgement (the vocabulary calibration and the realness head use the
-same padding), so a batch of 16 sounds costs about one batch of CLAP windows. Components:
+same padding), so a batch of 16 sounds costs about one batch of CLAP windows. Model components (v1):
match_z calibrated z of the description among earshot's vocabulary (strong >= 3, as `earshot match`)
alt_margin match_z minus the best alternative's z (must be > 0: it sounds more like the target than the confusions)
p_real earshot realness (clip thresholds: real >= 0.77; the pure-tone rule applies)
ref_typ_z how close it sits to the nearest real references compared with how close the references sit to each other
feat_pen how far its measurements (attack, decay, brightness, length, low end, bandwidth) fall outside the
references' spread (robust z beyond 1.5, averaged)
- total min(match_z, 6) + 2 * (2 * p_real - 1) + 0.5 * clip(ref_typ_z, -3, 2) - 0.7 * feat_pen
- - max(0, 0.5 - alt_margin)
- passes match_z >= 3 and alt_margin > 0 and p_real >= 0.77 and ref_typ_z >= -2 and feat_pen <= 1
-Optimising against earshot can find sounds that please its models without sounding better (Goodhart). `holdout()`
-uses the AudioSet tagger, which is not in the objective, but it stays held out only while nobody picks or steers by
-its tags (in the forge test it did steer four of five sounds, which biased their tags). Once a search has run, the
-model checks saturate: held-out phrasings and independent real sets agree too. The model-free `tone` check below is
-the signal that still separates made from real. The realness
-model in particular is easy to fool: modal resonators plus a room and a mic noise floor read "real" (p_real > 0.9)
-whether or not they sound like the target, so trust reference closeness, the held-out tags and spectrograms more.
-
-Fair comparison with the references: `score_refs()` scores every reference leave-one-out (its typicality and the
-feature spread are computed from the OTHER references), so a real recording is not its own nearest neighbour. With
-fewer than 3 other references there is no spread to compare with, and ref_typ_z is None. Each row also carries
-`alt_zs` (the z of every alternative, so you can see which confusion is closest). `extra=fn(x, row) -> float` adds a
-recipe-specific check (a model-free distance to the refs, say) to the total; its value is kept in row['extra'].
-
-Model-free ears (`tone=`): every row also carries row['tone'] from earshot.forge.tone (1/3-octave tonal balance against
-the refs' median, which bands fall outside every ref, the tail's t40), measured from the samples, no model. The model
-checks above saturate once a search has tuned against them (made sounds land inside or above the real recordings on
-every one, held-out ones included), while these measurements still show the gap. tone=W (default 0, the scores the
-forge test reported) subtracts W * ToneRef.penalty() from the total (0 while the sound is no farther from the refs'
-median balance than a typical real recording, then linear) and, with W > 0, `passes` also needs tone_db no larger
-than the farthest real reference's. tone_target= sets that "typical" distance in dB (and the pass limit to 1.6x it,
-the median : farthest proportion of the six real key presses) when the refs are too few or too different from each
-other for their own spread to mean anything (the four kit pops sit 15-24 dB apart).
+Model-free components (v2, measured from the samples; each its own number in the row):
+ tone earshot.forge.tone.ToneRef: 1/3-octave balance vs the refs' median (tone_db, tone_ratio, tone_pen)
+ env earshot.forge.tone.EnvRef: the spectro-temporal envelope (28 band trajectories in 10 ms frames) against
+ the nearest ref after onset / peak alignment with a small stretch / shift tolerance (env_db, env_ratio),
+ plus the cells (segment x band group) beyond every ref (env_out); env_pen = EnvRef.penalty
+ artifacts earshot.forge.polish: clicks, edge jumps, aliasing, DC / infrasound, harsh narrow resonances, comb ring,
+ truncated tail, digital silence inside, clipping, frozen spectrum (art_pen; `hard` lists the defects
+ that fail `passes`)
+ polish attack crispness, onset snap, tail smoothness, tail length, body / crest vs the refs (polish_pen)
+Modes: 'real' (refs are real recordings: realness counts) or 'designed' (refs are designed SFX or missing: glitch,
+sub drop, UI blips, risers; realness is reported but not required, clicks / digital silence are soft). mode='auto'
+picks 'designed' when there are no refs or the target names a designed category (DESIGNED_WORDS).
+
+Totals. version=1 reproduces the forge test's critic exactly:
+ total = min(match_z, 6) + 2 * (2 * p_real - 1) + 0.5 * clip(ref_typ_z, -3, 2) - 0.7 * feat_pen - max(0, 0.5 - margin)
+ [- tone * tone_pen when tone > 0]
+ passes = match_z >= 3 and margin > 0 and p_real >= 0.77 and ref_typ_z >= -2 and feat_pen <= 1 [and the tone rule]
+version=2 (default) keeps those terms (weights W2, realness weight 0 in designed mode) and subtracts the model-free
+penalties; why each weight:
+ match 1, real 2, typ 0.5, feat 0.7, alt 1 unchanged from v1 (calibrated in the forge test; the model checks saturate
+ after a search, so they mostly decide early rounds)
+ tone 1.0 always on in v2 (v1 recommended 1.5 when it was the only model-free check; env now carries part of the
+ same evidence, band levels over time include the average balance, so 1.0 avoids counting it twice)
+ env 1.0 one point per typical ref-to-ref spread of excess (env_ratio - 1) plus 1 per 6 dB beyond every ref; the
+ same scale as tone. The shape over time is what the blind test heard (the forged impact's tail)
+ art 1.0 each detector is capped at 2, so one blatant defect costs as much as the whole realness term; hard
+ defects also fail `passes` (a click or a clipped peak disqualifies a library sound)
+ polish 0.75 relative to the refs' ranges, which 3-6 refs pin down loosely: a little less than the artifacts
+ passes (v2) = the v1 model rules (realness only in real mode) and no hard artifact and env_pen <= 1.5 and
+ polish_pen <= 1.5 and tone_db <= 1.25 x the farthest ref's (or 2 x tone_target)
+`weights=` overrides any W2 entry (weights={'tone': 0} to drop the tone term); tone= > 0 sets the tone weight (v1
+callers keep their meaning).
+
+Fair comparison with the references: `score_refs()` scores every reference leave-one-out (its typicality, feature
+spread, tone, envelope and polish ranges are computed from the OTHER references), so a real recording is not its own
+nearest neighbour. With fewer than 3 other references there is no spread to compare with, and ref_typ_z / env_ratio
+are None. Each row also carries `alt_zs` (the z of every alternative). `extra=fn(x, row) -> float` adds a
+recipe-specific check to the total; its value is kept in row['extra'].
+
+Honest limits: optimising against earshot finds sounds that please its models without sounding better (Goodhart);
+the realness model is easy to fool (modal resonators plus a room and a mic floor read "real"). Nearest-ref distances
+(typicality, env_db) flatter averages: with few, diverse refs a sound in their middle sits closer to each than they
+sit to each other, so the v1 finals score env_ratio < 1 for pops, whooshes and impacts. The 'beyond every ref' cells,
+the artifact detectors and the polish ranges are what still separate made from real there. `holdout()` uses the
+AudioSet tagger, held out only while nobody picks or steers by its tags.
"""
import os
import numpy as np
from .. import audio, compare, features, realness, vocab
-from .tone import ToneRef
+from . import polish
+from .tone import EnvRef, ToneRef, bandgram
FEATS = ('attack_ms', 'decay_ms', 'brightness_hz', 'length_s', 'low_end_db', 'bandwidth_hz')
LOG = ('attack_ms', 'decay_ms', 'brightness_hz', 'length_s', 'bandwidth_hz')
+W2 = {'match': 1.0, 'real': 2.0, 'typ': 0.5, 'feat': 0.7, 'alt': 1.0, 'tone': 1.0, 'env': 1.0, 'art': 1.0,
+ 'polish': 0.75}
+DESIGNED_WORDS = ('glitch', 'stutter', 'sub drop', 'sub-drop', 'subdrop', 'blip', 'beep', 'bleep', 'synth', 'riser',
+ 'downlifter', 'uplifter', 'stinger', 'logo sting', 'braam', 'laser', 'zap', 'notification', 'shimmer',
+ 'bitcrush', 'digital')
+ENV_PASS, POLISH_PASS, TONE_PASS = 1.5, 1.5, 1.25
def _mono(x):
@@ -66,26 +94,57 @@ def _fv(k, v):
return np.log(max(v, 1e-3)) if k in LOG else float(v)
+def resolve_mode(mode, target, refs):
+ """'real' / 'designed' from mode='auto': designed when there are no refs or the target names a designed category"""
+ if mode in ('real', 'designed'):
+ return mode
+ t = (target or '').lower()
+ return 'designed' if not refs or any(w in t for w in DESIGNED_WORDS) else 'real'
+
+
+def total_v2(r, W=None, mode='real'):
+ """the v2 total and pass verdict from a row's components (pure; rows from Critic.score carry every input)"""
+ W = {**W2, **(W or {})}
+ mz, margin, p, typ = r['match_z'], r['alt_margin'], r['p_real'], r.get('ref_typ_z')
+ real = W['real'] if mode == 'real' else 0.0
+ total = (W['match'] * min(mz, 6.0) + real * (2 * p - 1) + (W['typ'] * float(np.clip(typ, -3, 2)) if typ is not None
+ else 0.0) - W['feat'] * r['feat_pen'] - W['alt'] * max(0.0, 0.5 - margin)
+ - W['tone'] * r.get('tone_pen', 0.0) - W['env'] * r.get('env_pen', 0.0)
+ - W['art'] * r.get('art_pen', 0.0) - W['polish'] * r.get('polish_pen', 0.0))
+ passes = bool(mz >= 3 and margin > 0 and (mode != 'real' or p >= 0.77) and (typ is None or typ >= -2)
+ and r['feat_pen'] <= 1.0 and not r.get('hard') and r.get('env_pen', 0.0) <= ENV_PASS
+ and r.get('polish_pen', 0.0) <= POLISH_PASS and not r.get('tone_far', False))
+ return float(total), passes
+
+
class Critic:
- def __init__(self, target, alts=(), refs=None, k=3, extra=None, tone=0.0, tone_target=None):
+ def __init__(self, target, alts=(), refs=None, k=3, extra=None, tone=0.0, tone_target=None, mode='auto',
+ version=2, weights=None):
self.target, self.alts = target, list(alts or [])
self.extra = extra
self.tone, self.tone_target = float(tone), tone_target
+ self.version = int(version)
+ self.W = {**W2, **(weights or {})}
+ if self.tone > 0 and not (weights and 'tone' in weights):
+ self.W['tone'] = self.tone
self.v = vocab.get()
self.Q = self.v.embed_text([target] + self.alts)
self.m = realness.load_model()
self.k = k
self.refs = audio.list_media(refs, exts=audio.AUDIO_EXTS) if refs else []
- self.A_r, self.ref_feats, self.ref_stats, self.tone_ref = None, [], {}, None
+ self.mode = resolve_mode(mode, target, self.refs)
+ self.A_r, self.ref_feats, self.ref_stats, self.tone_ref, self.env_ref = None, [], {}, None, None
+ xs = []
if self.refs:
self.A_r = vocab.embed_files(self.refs)
- xs = []
for p in self.refs:
x = audio.load(p)
xs.append(x)
self.ref_feats.append(compare.feats(features.measure(x, light=True), x))
self.ref_stats = self._stats(self.ref_feats)
self.tone_ref = ToneRef(xs)
+ self.env_ref = EnvRef(xs, names=self.refs)
+ self.polish_ref = polish.PolishRef(xs, designed=self.mode == 'designed')
@staticmethod
def _stats(ref_feats):
@@ -114,7 +173,7 @@ class Critic:
for i, x in enumerate(xs):
ex = exclude[i]
if ex is None:
- A_r, refs, stats, min_refs = self.A_r, self.refs, self.ref_stats, 2
+ A_r, refs, stats, min_refs = self.A_r, self.refs, self.ref_stats, 3 # 2 refs: one pair, no spread
else:
if ex not in loo:
keep = [j for j in range(len(self.refs)) if j != ex]
@@ -159,7 +218,7 @@ class Critic:
alt_margin=round(margin, 2), p_real=round(float(p), 3),
ref_typ_z=None if typ is None else round(float(typ), 2), nearest_ref=near,
feat_pen=round(pen, 2), feat_dev=dev, feats={k: fx.get(k) for k in FEATS if fx.get(k) is not None},
- total=round(float(total), 3), passes=passes, sounds_like=[t[0] for t in top])
+ total=round(float(total), 3), passes=passes, sounds_like=[t[0] for t in top], mode=self.mode)
if ex is not None:
row['left_out'] = os.path.basename(self.refs[ex])
if self.tone_ref is not None and self.tone_ref.n:
@@ -169,15 +228,21 @@ class Critic:
row['tone'] = {k: tm[k] for k in ('tone_db', 'tone_ratio', 'outside', 't40_ms', 't40_ref', 'length_s',
'length_ref', 'shares', 'shares_ref_med', 'words')}
row['tone_db'], row['tone_ratio'] = tm['tone_db'], tm['tone_ratio']
- if self.tone > 0:
- tpen = self.tone_ref.penalty(tm)
- row['tone_pen'] = round(tpen, 3)
+ tpen = self.tone_ref.penalty(tm)
+ row['tone_pen'] = round(tpen, 3)
+ far = max(tm['tone_ref']) if tm['tone_ref'] else None
+ if self.tone_target is not None: # median : farthest of a typical real set is ~1 : 1.6
+ far = 1.6 * float(self.tone_target)
+ if self.version == 1 and self.tone > 0:
row['total'] = round(row['total'] - self.tone * tpen, 3)
- far = max(tm['tone_ref']) if tm['tone_ref'] else None
- if self.tone_target is not None: # median : farthest of a typical real set is ~1 : 1.6
- far = 1.6 * float(self.tone_target)
if far is not None and tm['tone_db'] > far:
row['passes'] = False
+ row['tone_far'] = bool(far is not None and tm['tone_db'] > TONE_PASS * far)
+ self._model_free(x, row, ex)
+ row['total_v1'] = row['total']
+ if self.version >= 2:
+ t2, p2 = total_v2(row, self.W, self.mode)
+ row['total'], row['passes'] = round(t2, 3), p2
if self.extra is not None:
adj = float(self.extra(x, row))
row['extra'] = round(adj, 3)
@@ -186,9 +251,27 @@ class Critic:
rows.append(row)
return rows
+ def _model_free(self, x, row, ex):
+ """the v2 measurements: spectro-temporal envelope, artifacts and polish (in place on the row)"""
+ G = bandgram(x)
+ if self.env_ref is not None and self.env_ref.n:
+ em = self.env_ref.check(x, exclude=ex, G=G)
+ if em:
+ row['env'] = {k: em[k] for k in ('env_db', 'env_ratio', 'env_ref', 'env_out', 'env_out_ref', 'nearest',
+ 'anchor', 'stretch', 'shift_ms', 'outside', 'words')}
+ row['env_db'], row['env_ratio'], row['env_out'] = em['env_db'], em['env_ratio'], em['env_out']
+ row['env_pen'] = round(EnvRef.penalty(em), 3)
+ pm = self.polish_ref.check(x, exclude=ex, G=G)
+ row['artifacts'] = {'pen': pm['art'], 'hard': pm['hard'], 'limits': {k: v for k, v in pm['limits'].items()
+ if k in pm['art']}}
+ row['polish'] = {'pen': pm['polish'], 'limits': {k: v for k, v in pm['limits'].items() if k in pm['polish']}}
+ row['measures'] = {k: v for k, v in pm['measures'].items() if k not in ('res_peaks',)}
+ row['art_pen'], row['polish_pen'], row['hard'] = pm['art_pen'], pm['polish_pen'], pm['hard']
+ row['polish_words'] = pm['words']
+
def _summary(self, r):
bits = [f"match z {r['match_z']:+.2f}" + (f" (best alternative {r['alt_z']:+.2f})" if r['alt_z'] is not None else ''),
- f"p_real {r['p_real']:.2f}"]
+ f"p_real {r['p_real']:.2f}" + (' (not required: designed)' if r.get('mode') == 'designed' else '')]
if r['ref_typ_z'] is not None:
bits.append(f"typicality vs refs z {r['ref_typ_z']:+.2f} (nearest {r['nearest_ref']})")
off = [f'{k} {v:+.1f}' for k, v in r['feat_dev'].items() if abs(v) > 1.5]
@@ -197,6 +280,10 @@ class Critic:
bits.append('sounds like: ' + ', '.join(r['sounds_like']))
if r.get('tone'):
bits.append(r['tone']['words'])
+ if r.get('env'):
+ bits.append(r['env']['words'])
+ if r.get('polish_words'):
+ bits.append(r['polish_words'])
return ('PASS ' if r['passes'] else '') + f"total {r['total']:.2f}: " + '; '.join(bits)
def score_files(self, paths):
@@ -208,6 +295,14 @@ class Critic:
xs = [audio.load(p) for p in self.refs]
return self.score(xs, exclude=list(range(len(xs))) if loo else None)
+ def consistency(self, xs):
+ """loudness consistency of variants meant to be swapped in an edit: momentary-max LUFS per variant at its own
+ level, spread and std (polish.loudness_spread), with the refs' own spread when there are refs"""
+ out = polish.loudness_spread(xs)
+ if self.refs:
+ out['refs'] = polish.loudness_spread(self.refs)
+ return out
+
def holdout(self, xs):
"""checks that are NOT in the objective: AudioSet top labels and its synth-class probability (held out only
while they decide nothing: fix the selection rule before reading them)"""
diff --git a/earshot/forge/dsp.py b/earshot/forge/dsp.py
index 196b1eb..61ca783 100644
--- a/earshot/forge/dsp.py
+++ b/earshot/forge/dsp.py
@@ -7,6 +7,14 @@ as filtered turbulence, every recording has a room and a microphone. Recipes in
x = dsp.excite(hit, dsp.contact(0.0015, r)) # struck for 1.5 ms
x = dsp.mic(dsp.room(x, r, rt60=0.18, wet=0.08), r) # in a room, through a mic
+Designed sources for motion design (v2), all deterministic for a given rng:
+ boom = dsp.sub_drop(1.6, r, f0=130, f1=32) # pitch-dropping sine + harmonics + click
+ ding = dsp.bell(1320, 1.2, r, 'bar', decay=0.5) # inharmonic partials, doublet beating; or dsp.fm(...)
+ rise = dsp.shepard(3.0, r, up=True, kind='noise') # endless riser; dsp.sweep + dsp.osc for pitch sweeps
+ gust = dsp.air(1.0, r, center=900, flutter=0.6) # 1/f-ish turbulence with gusts and an optional whistle
+ hit = dsp.stack([(boom, 0, 0), (ding, 0.0, -9, 0.3)], onset_db=-24, peak_db=-1) # layers: offsets, gains, pans
+Effects (reverb, saturation, transient shaping, limiting, EQ match, Doppler, glitch, grains) live in fx.py.
+
No recorded audio is used anywhere: every sample is computed.
"""
import numpy as np
@@ -267,3 +275,285 @@ def trim_tail(x, floor_db=-70.0, keep=0.01):
def normalize(x, peak_db=-1.0):
return x * (10 ** (peak_db / 20) / (np.abs(x).max() + 1e-12))
+
+
+# --------------------------------------------------------------------------------------------- v2: curves + oscillators
+def curve(spec, n, shape='lin'):
+ """a per-sample control signal of length n from a scalar, an array (resampled to n if its length differs) or
+ breakpoints [(t_seconds, value), ...] ('lin' or 'exp' interpolation, see env_curve).
+ cut = dsp.curve([(0, 200), (1.5, 8000)], dsp.n_of(1.5), 'exp')"""
+ if np.isscalar(spec):
+ return np.full(n, float(spec))
+ if isinstance(spec, (list, tuple)) and len(spec) and isinstance(spec[0], (list, tuple)):
+ return env_curve([tuple(p) for p in spec], n / SR, shape)[:n]
+ a = np.asarray(spec, np.float64).ravel()
+ if len(a) == n:
+ return a
+ return np.interp(np.linspace(0, len(a) - 1, n), np.arange(len(a)), a)
+
+
+def sweep(f0, f1, seconds, curve='exp', power=1.0):
+ """a frequency trajectory (Hz per sample) from f0 to f1: 'exp' (equal octaves per second: the musical riser),
+ 'lin', or 'ease' (exponential with a power-law time warp: power > 1 starts slow and accelerates, the classic
+ tension riser; power < 1 moves fast first and then settles, a downlifter).
+ f = dsp.sweep(200, 6000, 3.0, 'ease', power=2.2); y = dsp.osc(f, 'saw')"""
+ n = n_of(seconds)
+ u = np.linspace(0, 1, n)
+ if curve == 'lin':
+ return f0 + (f1 - f0) * u
+ if curve == 'ease':
+ u = u ** power
+ return f0 * (f1 / f0) ** u
+
+
+@njit(cache=True)
+def _blep(t, dt):
+ if t < dt:
+ t = t / dt
+ return t + t - t * t - 1.0
+ if t > 1.0 - dt:
+ t = (t - 1.0) / dt
+ return t * t + t + t + 1.0
+ return 0.0
+
+
+@njit(cache=True)
+def _osc(freq, shape, ph):
+ n = len(freq)
+ y = np.empty(n)
+ for i in range(n):
+ dt = min(max(freq[i] / 48000.0, 1e-9), 0.49)
+ if shape == 0:
+ y[i] = np.sin(2 * np.pi * ph)
+ elif shape == 1:
+ y[i] = 2.0 * ph - 1.0 - _blep(ph, dt)
+ else:
+ v = 1.0 if ph < 0.5 else -1.0
+ y[i] = v + _blep(ph, dt) - _blep((ph + 0.5) % 1.0, dt)
+ ph += dt
+ if ph >= 1.0:
+ ph -= 1.0
+ return y
+
+
+def osc(freq, shape='sine', seconds=None, phase=0.0):
+ """band-limited oscillator (PolyBLEP saw / square, so sweeps up to 10 kHz do not alias): freq = Hz scalar (with
+ seconds) or a per-sample trajectory (dsp.sweep); shape 'sine' | 'saw' | 'square' | 'tri'; peak about 1.
+ y = dsp.osc(dsp.sweep(110, 880, 2.0), 'saw')"""
+ f = np.asarray(freq, np.float64)
+ if f.ndim == 0:
+ f = np.full(n_of(seconds or 1.0), float(f))
+ k = {'sine': 0, 'saw': 1, 'square': 2, 'tri': 3}[shape]
+ y = _osc(np.ascontiguousarray(f), 2 if k == 3 else k, float(phase) % 1.0)
+ if k == 3: # integrate the band-limited square (leaky, from -1): a triangle, harmonics -12 dB/oct
+ y, _ = signal.lfilter([1.0], [1.0, -0.9995], y * 4 * f / SR, zi=[-0.9995 * (1 - 2 * (float(phase) % 1.0))])
+ return y
+
+
+def shepard(seconds, r, up=True, rate=0.5, f_lo=40.0, f_hi=10000.0, kind='tone', q=6.0, tilt=-1.5):
+ """Shepard / Risset glissando: voices an octave apart glide forever (rate octaves per second, up or down) under a
+ raised-cosine loudness bell over log-frequency, so the pitch seems to rise (or fall) without end: the endless
+ riser / downlifter. kind 'tone' (sines) or 'noise' (resonant noise bands, q), tilt dB per octave. Unit RMS.
+ rise = dsp.shepard(4.0, dsp.rng(1), up=True, rate=0.6, kind='noise')"""
+ n = n_of(seconds)
+ t = t_of(n)
+ octs = np.log2(f_hi / f_lo)
+ v = max(2, int(round(octs)))
+ out = np.zeros(n)
+ sgn = 1.0 if up else -1.0
+ for k in range(v):
+ u = (k / v + sgn * rate * t / octs) % 1.0
+ f = f_lo * 2 ** (u * octs)
+ a = (0.5 - 0.5 * np.cos(2 * np.pi * u)) * 10 ** (tilt * u * octs / 20)
+ if kind == 'tone':
+ out += a * np.sin(2 * np.pi * np.cumsum(f) / SR + r.uniform(0, 2 * np.pi))
+ else:
+ out += a * svf(r.standard_normal(n), f, q, 'bandn') * np.sqrt(q)
+ return out / (np.sqrt(np.mean(out ** 2)) + 1e-12)
+
+
+# ------------------------------------------------------------------------------------------------ v2: air
+def wander(seconds, r, rate=4.0, octaves=3):
+ """a smooth 1/f-ish random control signal, unit std, mean 0: low-passed noises at rate, rate/3, rate/9 Hz with
+ rising weights (slow drift dominates, fast flutter on top). For gusts, drifting pitch, moving pans.
+ c = dsp.wander(2.0, dsp.rng(5), rate=6)"""
+ n = n_of(seconds)
+ m = max(16, int(n / SR * 1000) + 4) # computed at 1 kHz, then interpolated
+ c = np.zeros(m)
+ for i in range(octaves):
+ fr = float(np.clip(rate / 3 ** i, 0.05, 400))
+ pad = int(2000 / fr) + 8
+ w = r.standard_normal(m + 2 * pad)
+ sos = signal.butter(2, fr / 500, output='sos')
+ c += signal.sosfiltfilt(sos, w)[pad:pad + m] / np.sqrt(fr) * 0.8 ** i
+ c = np.interp(np.arange(n) / SR * 1000, np.arange(m), c)
+ return (c - c.mean()) / (c.std() + 1e-12)
+
+
+def air(seconds, r, center=1200.0, lo_db=6.0, hi_db=4.5, flutter=0.5, rate=6.0, bright=0.6, whistle=0.0):
+ """moving-air turbulence: noise with a broad aerodynamic spectrum (its density rises lo_db per octave below
+ `center` and falls hi_db per octave above: 3-5 is the 1/f-ish slope of real airflow; 1/3-octave band levels fall
+ 3 dB per octave less), gusting `flutter` (0..1) at about `rate` Hz
+ with a 1/f wander, louder gusts also brighter (bright 0..1: faster air moves the spectrum up), and an optional
+ `whistle` (0..1): a narrow resonance that wanders with the flow. Unit RMS; shape it with an envelope.
+ w = dsp.air(1.2, dsp.rng(3), center=900, flutter=0.6) * dsp.env_curve([(0, 0), (.5, 1), (1.2, 0)], 1.2)"""
+ n = n_of(seconds)
+ F = np.fft.rfft(r.standard_normal(n))
+ f = np.fft.rfftfreq(n, 1 / SR)
+ u = np.log2(np.maximum(f, 5.0) / center)
+ F *= 10 ** (-10 * np.log10(2 ** (-lo_db / 3.0103 * u) + 2 ** (hi_db / 3.0103 * u)) / 20)
+ x = np.fft.irfft(F, n)
+ x /= np.sqrt(np.mean(x ** 2)) + 1e-12
+ c = wander(seconds, r, rate)[:n]
+ if bright > 0:
+ x = svf(x, np.clip(center * 6 * 2 ** (1.2 * bright * flutter * c), 200, 20000), 0.6, 'low')
+ if whistle > 0:
+ c2 = 0.6 * c + 0.8 * wander(seconds, r, rate * 0.4)[:n]
+ wh = svf(r.standard_normal(n), center * 2 ** (0.5 * c2), 9.0, 'bandn')
+ x = x / (np.sqrt(np.mean(x ** 2)) + 1e-12) + whistle * wh / (np.sqrt(np.mean(wh ** 2)) + 1e-12)
+ x = x * np.exp(1.1 * flutter * c)
+ return x / (np.sqrt(np.mean(x ** 2)) + 1e-12)
+
+
+# ------------------------------------------------------------------------------------------------ v2: designed sources
+def sub_drop(seconds=1.6, r=None, f0=120.0, f1=34.0, drop=0.25, decay=0.6, harm=0.3, click=0.25, attack=0.004):
+ """sub drop / boom: a sine whose pitch falls exponentially from f0 to f1 (time constant `drop` s), harmonics 2 and
+ 3 at `harm` (so it still reads on laptop speakers), a short bright `click` on the onset (0 = none), amplitude
+ decaying with time constant `decay` s and a clean fade. Peak -1 dBFS.
+ boom = dsp.sub_drop(1.8, dsp.rng(2), f0=140, f1=30, drop=0.3)"""
+ r = r if r is not None else rng(0)
+ n = n_of(seconds)
+ t = t_of(n)
+ f = f1 + (f0 - f1) * np.exp(-t / max(drop, 1e-3))
+ ph = 2 * np.pi * np.cumsum(f) / SR
+ y = np.sin(ph) + harm * (0.55 * np.sin(2 * ph + 0.3) + 0.3 * np.sin(3 * ph + 0.7))
+ e = np.exp(-t / max(decay, 1e-3))
+ a = min(n, n_of(attack))
+ e[:a] *= 0.5 - 0.5 * np.cos(np.linspace(0, np.pi, a))
+ y *= e
+ if click > 0:
+ k = min(n, n_of(0.012))
+ tk = t_of(k)
+ ck = butter(r.standard_normal(k), 'high', 1800) * np.exp(-tk / 0.0012)
+ ck += 0.6 * np.sin(2 * np.pi * 2400 * tk) * np.exp(-tk / 0.003)
+ y[:k] += click * ck / (np.abs(ck).max() + 1e-12)
+ return normalize(fade(y, 0.0, seconds * 0.3), -1.0) # a long fade: a sub must not stop audibly
+
+
+BELLS = { # partial ratios and amplitudes: church bell (hum, prime, tierce, quint, nominal, ...), tubular chime,
+ # glockenspiel bar (free-free beam), wine glass, singing bowl
+ 'church': ([0.5, 1.0, 1.183, 1.506, 2.0, 2.514, 2.662, 3.011, 4.166, 5.433, 6.796, 8.215],
+ [0.55, 0.8, 0.7, 0.35, 1.0, 0.45, 0.4, 0.3, 0.25, 0.18, 0.12, 0.08]),
+ 'tubular': ([1.0, 2.756, 5.404, 8.933, 13.34], [0.35, 1.0, 0.7, 0.4, 0.2]),
+ 'bar': ([1.0, 2.756, 5.404, 8.933], [1.0, 0.3, 0.12, 0.06]),
+ 'glass': ([1.0, 2.32, 4.25, 6.63, 9.38], [1.0, 0.45, 0.25, 0.12, 0.06]),
+ 'bowl': ([1.0, 2.71, 5.13, 8.27], [1.0, 0.6, 0.3, 0.12]),
+}
+
+
+def bell(f0, seconds, r, partials='bar', decay=1.2, slope=0.8, bright=0.5, beat=1.0, stretch=0.0, strike=0.3):
+ """additive bell / chime / ding with inharmonic partials (a BELLS preset name or (ratios, amps)): each partial
+ decays with time constant decay * ratio^-slope (highs die first, as in real metal and glass) and is a doublet
+ beating at about `beat` Hz (the slow shimmer of a real bell); bright 0..1 lifts the upper partials, stretch bends
+ the ratios (ratio^(1+stretch)), strike adds a short mallet transient. Peak -1 dBFS.
+ ding = dsp.bell(1320, 1.5, dsp.rng(4), 'bar', decay=0.5, bright=0.3)"""
+ ratios, amps = BELLS[partials] if isinstance(partials, str) else partials
+ n = n_of(seconds)
+ t = t_of(n)
+ y = np.zeros(n)
+ for q, a in zip(ratios, amps):
+ q = q ** (1 + stretch)
+ f = f0 * q
+ if f > SR * 0.45:
+ continue
+ a = a * q ** (1.2 * (bright - 0.5))
+ b = beat * r.uniform(0.4, 1.6) * (1 if q >= 1 else 0.5)
+ env = a * np.exp(-t / max(decay * q ** -slope, 1e-4))
+ y += env * (np.sin(2 * np.pi * f * t + r.uniform(0, 6.3)) +
+ 0.7 * np.sin(2 * np.pi * (f + b) * t + r.uniform(0, 6.3))) / 1.7
+ a = min(n, n_of(0.0015))
+ y[:a] *= np.linspace(0, 1, a)
+ if strike > 0:
+ k = min(n, n_of(0.006))
+ s = butter(r.standard_normal(k), 'band', (f0 * 2, min(f0 * 12, 18000))) * np.exp(-t_of(k) / 0.0008)
+ y[:k] += strike * s * np.abs(y).max() / (np.abs(s).max() + 1e-12)
+ return normalize(fade(y, 0.0, seconds * 0.15), -1.0)
+
+
+def fm(fc, ratio, index, seconds, decay=1.0, index_decay=0.3, attack=0.002):
+ """two-operator FM (Chowning): carrier fc, modulator fc*ratio (non-integer ratios ring inharmonic: 1.4 is the
+ classic bell, 3.5 a metallic ding), modulation index decaying from `index` with time constant index_decay (bright
+ strike, pure tail), amplitude time constant `decay`. Peak about 1.
+ ding = dsp.fm(880, 1.4, 6, 1.2, decay=0.4, index_decay=0.08)"""
+ n = n_of(seconds)
+ t = t_of(n)
+ I = index * np.exp(-t / max(index_decay, 1e-4))
+ y = np.sin(2 * np.pi * fc * t + I * np.sin(2 * np.pi * fc * ratio * t)) * np.exp(-t / max(decay, 1e-4))
+ a = min(n, n_of(attack))
+ y[:a] *= 0.5 - 0.5 * np.cos(np.linspace(0, np.pi, a))
+ return fade(y, 0.0, seconds * 0.15)
+
+
+# ------------------------------------------------------------------------------------------------ v2: layering
+def onset(x, db=-24.0):
+ """seconds until |x| (mono or stereo) first reaches `db` under its peak: where a layer 'hits'"""
+ m = np.abs(np.asarray(x, np.float64))
+ m = m.max(axis=1) if m.ndim == 2 else m
+ if not m.size or m.max() <= 0:
+ return 0.0
+ return float(np.argmax(m >= m.max() * 10 ** (db / 20))) / SR
+
+
+def align(x, t=0.0, db=-24.0):
+ """shift x (mono or stereo) so its onset (see onset) lands at t seconds: leading samples are trimmed or padded"""
+ x = np.asarray(x, np.float64)
+ d = n_of(t) * (t > 0) - int(round(onset(x, db) * SR))
+ if d >= 0:
+ return np.concatenate([np.zeros((d,) + x.shape[1:]), x])
+ return x[-d:]
+
+
+def pan_to_stereo(x, pan=0.0):
+ """mono -> stereo with constant-power pan (-1 left .. 1 right; centre keeps the level of each channel = x);
+ a stereo x is balanced the same way"""
+ x = np.asarray(x, np.float64)
+ th = (np.clip(pan, -1, 1) + 1) * np.pi / 4
+ g = np.array([np.cos(th), np.sin(th)]) * np.sqrt(2)
+ return x[:, None] * g if x.ndim == 1 else x * g
+
+
+def stack(layers, length=None, onset_db=None, peak_db=None):
+ """mix layers: each a dict {'x', 't' (s, default 0), 'gain' (dB, default 0), 'pan' (-1..1, optional)} or a tuple
+ (x, t, gain_db[, pan]). onset_db (e.g. -24) aligns each layer's onset (not its first sample) to its t. Returns
+ stereo (n, 2) when any layer is stereo or panned, else mono; length in s (default: until the last layer ends);
+ peak_db normalises the mix (layers add up: two -1 dBFS layers can peak above 0 dBFS without it).
+ hit = dsp.stack([(sub, 0, 0), (crack, 0.0, -6, -0.2), {'x': tail, 't': 0.01, 'gain': -12, 'pan': 0.3}])"""
+ ls = []
+ for L in layers:
+ if not isinstance(L, dict):
+ L = dict(zip(('x', 't', 'gain', 'pan'), L))
+ x = np.asarray(L['x'], np.float64)
+ t = float(L.get('t') or 0.0)
+ if onset_db is not None:
+ t -= onset(x, onset_db)
+ if t < 0:
+ x, t = x[n_of(-t):], 0.0
+ ls.append((x, t, 10 ** (float(L.get('gain') or 0.0) / 20), L.get('pan')))
+ st = any(x.ndim == 2 or p is not None for x, _, _, p in ls)
+ n = n_of(length) if length else max(int(round(t * SR)) + len(x) for x, t, _, _ in ls)
+ out = np.zeros((n, 2) if st else n)
+ for x, t, g, p in ls:
+ place(out, pan_to_stereo(x, p or 0.0) if st and (x.ndim == 1 or p) else x, t, g)
+ return out if peak_db is None else normalize(out, peak_db)
+
+
+def xfade(a, b, seconds=0.02, power=True):
+ """a then b, overlapped by `seconds` with an equal-power (power=True: for unrelated material) or linear crossfade
+ (power=False: for correlated material, e.g. two takes of the same tone); both mono or both stereo"""
+ a, b = np.asarray(a, np.float64), np.asarray(b, np.float64)
+ k = min(n_of(seconds), len(a), len(b))
+ u = np.linspace(0, 1, k)
+ fo, fi = (np.cos(u * np.pi / 2), np.sin(u * np.pi / 2)) if power else (1 - u, u)
+ if a.ndim == 2:
+ fo, fi = fo[:, None], fi[:, None]
+ return np.concatenate([a[:len(a) - k], a[len(a) - k:] * fo + b[:k] * fi, b[k:]])
diff --git a/earshot/forge/fx.py b/earshot/forge/fx.py
new file mode 100644
index 0000000..16f1b8b
--- /dev/null
+++ b/earshot/forge/fx.py
@@ -0,0 +1,753 @@
+"""Effects for sounds written in code (48 kHz float, mono (n,) or stereo (n, 2)): space, dynamics, colour, motion,
+glitch. Everything is deterministic (randomness only through an explicit rng) and computed: no recorded audio.
+
+ from earshot.forge import dsp, fx
+ r = dsp.rng(7)
+ hit = dsp.sub_drop(1.2, r) # a computed source
+ y = fx.saturate(hit, drive_db=9, mode="tape") # colour, oversampled
+ y = fx.transient(y, attack_db=4, sustain_db=3) # punchier onset, longer tail
+ y = fx.reverb(y, r, rt60=1.8, size=0.7, predelay=0.012, wet=0.3, stereo=True) # smooth FDN hall
+ y = fx.limit(y, ceiling_db=-1.0) # never above -1 dBFS
+
+Space reverb (16-line modulated FDN, frequency-dependent decay, early reflections), early (reflections only)
+Dynamics saturate (tanh / tape / hard / fold, oversampled), transient (attack / sustain), limit (look-ahead)
+Colour split (Linkwitz-Riley bands that sum flat), match_curve / eq_bands / eq_match (fit a smooth 1/3-octave EQ
+ so the tonal balance moves toward a target: tone.profile bands, minimum- or linear-phase FIR)
+Motion doppler (pass-by with distance, air absorption, pan), pan / width / decorrelate (static or over time),
+ sweep_filter (resonant exponential sweeps for risers and downlifters), varispeed / tape_stop
+Glitch crush (bits + sample-rate), stutter, glitch (sliced repeat / reverse / drop / octave / crush), granular
+ (grain cloud over a computed source), reverse_swell (a computed tail reversed into a hit)
+"""
+import numpy as np
+from scipy import ndimage, signal
+
+from .dsp import SR, butter, curve, n_of, njit, normalize, place, svf
+
+
+def _2d(x):
+ x = np.asarray(x, np.float64)
+ return (x[:, None], True) if x.ndim == 1 else (x, False)
+
+
+def _back(y, mono):
+ return y[:, 0] if mono else y
+
+
+def _rms(x):
+ return float(np.sqrt(np.mean(np.asarray(x, np.float64) ** 2)) + 1e-12)
+
+
+def _primes(hi):
+ s = np.ones(hi + 1, bool)
+ s[:2] = False
+ for i in range(2, int(hi ** 0.5) + 1):
+ if s[i]:
+ s[i * i::i] = False
+ return np.nonzero(s)[0]
+
+
+# ------------------------------------------------------------------------------------------------ space
+@njit(cache=True)
+def _allpass(x, d, g):
+ y = np.empty_like(x)
+ buf = np.zeros(d)
+ j = 0
+ for i in range(len(x)):
+ v = buf[j]
+ u = x[i] + g * v
+ y[i] = v - g * u
+ buf[j] = u
+ j += 1
+ if j >= d:
+ j = 0
+ return y
+
+
+@njit(cache=True)
+def _fdn(xin, B, C, dl, depth, rate, phase, lb0, lb1, la1, hb0, hb1, ha1, n_out):
+ N = len(dl)
+ L = 1
+ while L < int(dl.max() + depth.max() + 8):
+ L *= 2
+ msk = L - 1
+ buf = np.zeros((N, L))
+ zl = np.zeros(N)
+ zh = np.zeros(N)
+ ya = np.zeros(N)
+ v = np.zeros(N)
+ out = np.zeros((n_out, C.shape[1]))
+ nin = xin.shape[0]
+ nc = xin.shape[1]
+ sc = 1.0 / np.sqrt(N)
+ w = 0
+ for i in range(n_out):
+ for k in range(N):
+ d = dl[k] + depth[k] * np.sin(2 * np.pi * rate[k] * i / 48000.0 + phase[k])
+ M = int(np.floor(d - 0.5)) # all-pass (Thiran) fractional read: unity gain at
+ fd = d - M # every frequency, so modulation does not darken
+ eta = (1.0 - fd) / (1.0 + fd)
+ y = eta * buf[k, (w - M) & msk] + buf[k, (w - M - 1) & msk] - eta * ya[k]
+ ya[k] = y
+ t = lb0[k] * y + zl[k] # low shelf (bass decay) then high shelf (damping)
+ zl[k] = lb1[k] * y - la1[k] * t
+ u = hb0[k] * t + zh[k]
+ zh[k] = hb1[k] * t - ha1[k] * u
+ v[k] = u
+ for c in range(C.shape[1]):
+ s = 0.0
+ for k in range(N):
+ s += C[k, c] * v[k]
+ out[i, c] = s
+ h = 1 # fast Walsh-Hadamard mix (orthogonal, dense)
+ while h < N:
+ for a in range(0, N, 2 * h):
+ for b in range(a, a + h):
+ p = v[b]
+ q = v[b + h]
+ v[b] = p + q
+ v[b + h] = p - q
+ h *= 2
+ for k in range(N):
+ s = v[k] * sc
+ if i < nin:
+ for c in range(nc):
+ s += B[c, k] * xin[i, c]
+ buf[k, w & msk] = s
+ w += 1
+ return out
+
+
+def _shelf(G, fc, high):
+ """first-order shelf coefficients (b0, b1, a1): DC gain G (low) or Nyquist gain G (high), the other end 1"""
+ K = np.tan(np.pi * np.clip(fc, 20, SR * 0.45) / SR)
+ s = np.sqrt(G)
+ if high:
+ b0, b1, a0, a1 = s + K, K - s, 1 / s + K, K - 1 / s
+ else:
+ b0, b1, a0, a1 = 1 + K * s, K * s - 1, 1 + K / s, K / s - 1
+ return b0 / a0, b1 / a0, a1 / a0
+
+
+def early(x, r, size=0.5, taps=18, damp=0.5, width=1.0):
+ """early reflections only: a sparse tapped delay per channel (3 ms .. 6+50*size ms, 1/distance amplitudes, later
+ taps darker), energy-normalised; stereo out (n + spread, 2).
+ er = fx.early(click, dsp.rng(1), size=0.4)"""
+ xm = np.asarray(x, np.float64)
+ xm = xm.mean(axis=1) if xm.ndim == 2 else xm
+ t_hi = 0.006 + 0.05 * size
+ m = n_of(t_hi) + 2
+ ir = np.zeros((m, 2))
+ for c in range(2):
+ tt = np.sort(np.exp(r.uniform(np.log(0.003), np.log(t_hi), taps)))
+ if c == 1:
+ tt = tt * (1 + width * r.uniform(-0.12, 0.12, taps))
+ for k, ti in enumerate(tt):
+ ir[min(m - 1, n_of(ti)), c] += r.choice([-1.0, 1.0]) * 0.003 / ti
+ for c in range(2): # distance darkening: split in thirds, darker later
+ thirds = np.array_split(np.arange(m), 3)
+ parts = []
+ for j, idx in enumerate(thirds):
+ seg = np.zeros(m)
+ seg[idx] = ir[idx, c]
+ parts.append(butter(seg, 'low', 14000 - damp * 4000 * (j + 1), order=1))
+ ir[:, c] = np.sum(parts, axis=0)
+ ir /= np.sqrt(np.sum(ir ** 2, axis=0, keepdims=True)) + 1e-12
+ return np.stack([signal.fftconvolve(xm, ir[:, c]) for c in range(2)], axis=1)
+
+
+def _late_noise(n, r, rt_f, f_axis, width):
+ """Polack's late field: Gaussian noise whose every frequency decays at its own rt60 (STFT-shaped), unit energy at
+ the median rt60; mid / side with the side high-passed at 200 Hz (the low end stays coherent, as in real rooms)"""
+ out = []
+ for _ in range(2):
+ f, t, Z = signal.stft(r.standard_normal(n + 1024), SR, nperseg=1024, noverlap=768)
+ rt = np.interp(f, f_axis, rt_f)
+ Z *= np.exp(-6.91 * t[None, :] / rt[:, None])
+ out.append(signal.istft(Z, SR, nperseg=1024, noverlap=768)[1][:n])
+ m, sd = out[0], butter(out[1], 'high', 200)
+ late = np.stack([m + width * sd, m - width * sd], axis=1) / np.sqrt(1 + width ** 2 * 0.9)
+ return late * np.sqrt(13.8 / (np.median(rt_f) * SR))
+
+
+def reverb_ir(r, rt60=1.6, size=0.6, predelay=0.01, damp=0.5, bass=1.2, mod=1.0, diffuse=0.7, early_db=-6.0,
+ width=1.0, late='noise', lines=16, tail=None, inputs=1, hf_hz=4000.0, lf_hz=250.0, air=1.0, lowcut=0.0):
+ """the impulse response reverb() convolves with, shape (n, inputs, 2); unit energy at mid frequencies (a darker
+ tail has less). late='noise': the FDN builds the onset and the first 0.1-0.2 s, then hands over (equal power, one
+ shared decay design) to a stochastic late field whose every frequency decays at its own rt60. A real room's late
+ reverb is Gaussian noise, so this tail never rings; a pure FDN (late='fdn') keeps resolvable modes in long tails
+ (long-FFT peaks 13-16 dB over the local mean at rt60 2.5-5 s, against 8-10 dB for noise). air scales the
+ absorption of air itself (about 0.17 dB per metre at 10 kHz, growing with f^2: a long tail darkens as it decays,
+ 10 kHz cannot ring longer than about 1 s); lowcut (Hz) high-passes the reverb (keeps a sub out of the room).
+ ir = fx.reverb_ir(dsp.rng(1), rt60=2.0)"""
+ nl = int(lines)
+ assert nl in (4, 8, 16, 32), 'lines must be 4, 8, 16 or 32 (Hadamard mixing)'
+ rt_lo, rt_hi = rt60 * bass, rt60 * max(0.05, 1 - 0.85 * damp)
+ tail = 1.2 * max(rt60, rt_lo) if tail is None else tail
+ pd = n_of(predelay) if predelay > 0 else 0
+ n = max(n_of(tail), 64)
+ lo = (0.006 + 0.024 * size) * SR # delays log-spread over [lo, 2.7 lo], distinct primes
+ tgt = np.exp(np.linspace(np.log(lo), np.log(2.7 * lo), nl)) * np.exp(r.uniform(-0.06, 0.06, nl))
+ pr = _primes(int(3.2 * lo) + 50)
+ dl, used = [], set()
+ for d in tgt:
+ p = next(int(c) for c in pr[np.argsort(np.abs(pr - d))] if int(c) not in used)
+ used.add(p)
+ dl.append(p)
+ dl = r.permutation(np.array(dl, np.float64))
+ g = lambda rt: 10 ** (-3 * dl / (max(rt, 0.02) * SR)) # per-pass gain for a decay rt (s)
+ gm, gl, gh = g(rt60), g(rt_lo), g(rt_hi)
+ lb0, lb1, la1 = _shelf(gl / gm, lf_hz, False)
+ hb0, hb1, ha1 = _shelf(gh / gm, hf_hz, True)
+ lb0, lb1 = lb0 * gm, lb1 * gm
+ ir = np.zeros((n, inputs, 2))
+ for c in range(inputs):
+ x = np.zeros(n)
+ x[0] = 1.0
+ if diffuse > 0: # input diffusion: 4 series all-passes
+ for j, d in enumerate((np.array([1.1, 1.7, 2.9, 4.3]) * (0.6 + 0.8 * size) * 1e-3 * SR).astype(int)):
+ x = _allpass(np.ascontiguousarray(x), int(d + 3 * c + j), min(0.8, diffuse))
+ B = r.choice([-1.0, 1.0], (1, nl)) / np.sqrt(nl)
+ Cm = r.choice([-1.0, 1.0], (nl, 2))
+ Cm[:, 1] -= Cm[:, 0] * (Cm[:, 0] @ Cm[:, 1]) / nl # Gram-Schmidt: decorrelated L / R taps
+ Cm /= np.linalg.norm(Cm, axis=0, keepdims=True)
+ depth = mod * (3 + 9 * r.uniform(0, 1, nl)) * (0.5 + 0.5 * size)
+ y = _fdn(np.ascontiguousarray(x[:, None]), B, np.ascontiguousarray(Cm), dl, depth, r.uniform(0.15, 1.1, nl),
+ r.uniform(0, 2 * np.pi, nl), lb0, lb1, la1, hb0, hb1, ha1, n)
+ y /= np.sqrt(np.mean(gm ** 2 / (1 - gm ** 2)) / nl) # unit energy: 1/N reaches the taps per pass
+ if late == 'noise':
+ fa = np.geomspace(20, SR / 2 * 0.999, 64) # the FDN's own decay per frequency, for the noise
+ k = int(np.argmin(np.abs(dl - np.median(dl))))
+ H = np.abs(signal.freqz([lb0[k], lb1[k]], [1, la1[k]], fa, fs=SR)[1] *
+ signal.freqz([hb0[k], hb1[k]], [1, ha1[k]], fa, fs=SR)[1])
+ rt_f = -3 * dl[k] / (SR * np.log10(np.minimum(H, 0.999999)))
+ tx = 0.06 + 0.1 * size
+ u = np.clip((np.arange(n) / SR - 0.5 * tx) / tx, 0, 1)
+ y = y * np.cos(u * np.pi / 2)[:, None] + _late_noise(n, r, rt_f, fa, 1.0) * np.sin(u * np.pi / 2)[:, None]
+ if early_db is not None and early_db > -90:
+ e = 10 ** (early_db / 10)
+ y /= np.sqrt(1 + e)
+ place(y, early(np.eye(1, n, 0)[0], r, size, damp=damp, width=width)[:n] * np.sqrt(e / (1 + e)), 0.0)
+ m, sd = (y[:, 0] + y[:, 1]) / 2, (y[:, 0] - y[:, 1]) / 2
+ ir[:, c] = np.stack([m + width * sd, m - width * sd], axis=1)
+ if air > 0: # air absorption: every bin decays a little more
+ a_db = air * 0.17 * 343.0 * (np.fft.rfftfreq(1024, 1 / SR) / 10000) ** 2 # dB per second
+ for c in range(inputs):
+ for ch in range(2):
+ f, t, Z = signal.stft(ir[:, c, ch], SR, nperseg=1024, noverlap=768, boundary='even')
+ Z *= 10 ** (-a_db[:, None] * np.maximum(t[None, :] - 512 / SR, 0) / 20)
+ ir[:, c, ch] = signal.istft(Z, SR, nperseg=1024, noverlap=768, boundary=True)[1][:n]
+ if lowcut:
+ ir = butter(ir, 'high', lowcut)
+ k = min(n, n_of(min(0.1, tail * 0.2)))
+ ir[-k:] *= (np.linspace(1, 0, k) ** 2)[:, None, None]
+ return np.concatenate([np.zeros((pd, inputs, 2)), ir])
+
+
+def reverb(x, r, rt60=1.6, size=0.6, predelay=0.01, damp=0.5, bass=1.2, mod=1.0, diffuse=0.7, early_db=-6.0,
+ width=1.0, wet=0.3, dry=1.0, late='noise', lines=16, tail=None, stereo=None, air=1.0, lowcut=0.0):
+ """a smooth algorithmic reverb: a 16-line feedback delay network (Hadamard mix, mutually prime delays scaled by
+ size 0..1, slowly modulated all-pass-interpolated taps) builds a dense onset, then a stochastic late field takes
+ over (see reverb_ir: it never rings metallic); decay rt60 s at mid frequencies, damp 0..1 shortens it above 4 kHz
+ (rt60 * (1 - 0.85 damp): soft walls) and air absorption (air, growing with f^2) darkens long tails further, bass
+ multiplies it below 250 Hz, lowcut (Hz) keeps the lows dry; an all-pass input diffuser (diffuse 0..0.8: smooth
+ onset, no flutter), early reflections at early_db under the tail, predelay s, stereo width 0..1 (the low end
+ stays coherent). The wet IR has unit energy, so wet is the reverb level relative to the direct
+ sound (0.3 = -10 dB). Returns len(x) + predelay + tail (default 1.2 * the longest rt60); mono in gives mono out
+ (dry + the wet's mid) unless stereo=True.
+ y = fx.reverb(hit, dsp.rng(1), rt60=2.4, size=0.8, damp=0.6, wet=0.35, stereo=True)"""
+ x2, mono = _2d(x)
+ stereo = (not mono) if stereo is None else stereo
+ ir = reverb_ir(r, rt60, size, predelay, damp, bass, mod, diffuse, early_db, width, late, lines, tail, x2.shape[1],
+ air=air, lowcut=lowcut)
+ n_out = len(x2) + len(ir) - 1
+ wetv = np.zeros((n_out, 2))
+ for c in range(x2.shape[1]):
+ wetv += signal.fftconvolve(x2[:, c:c + 1], ir[:, c], axes=0)
+ if not stereo: # mono: the mid, sqrt(2) keeps the wet energy
+ y = wet * wetv.mean(axis=1) * np.sqrt(2)
+ y[:len(x2)] += dry * x2.mean(axis=1)
+ return y
+ out = wet * wetv
+ out[:len(x2)] += dry * (np.repeat(x2, 2, axis=1) if x2.shape[1] == 1 else x2)
+ return out
+
+
+# ------------------------------------------------------------------------------------------------ dynamics + colour
+def saturate(x, drive_db=6.0, mode='tanh', mix=1.0, bias=0.0, os=None, level='rms'):
+ """saturation, oversampled `os` times (Kaiser polyphase; default 4, 8 for 'hard' and 'fold') so harmonics do
+ not alias (a 7 kHz sine driven 18 dB through tanh: aliases -53 dB under the harmonics, -14 dB without). Modes:
+ 'tanh' (soft clip), 'tape' (pre-emphasis, asymmetric soft clip with a little even-harmonic bias, de-emphasis,
+ gentle top roll-off), 'hard' (clip), 'fold' (sine wavefolder, aggressive). drive_db into the curve, mix dry/wet,
+ bias adds even harmonics; level 'rms' matches the input RMS, 'peak' the input peak, None leaves the curve's.
+ y = fx.saturate(sub, drive_db=12, mode='tape')"""
+ x2, mono = _2d(x)
+ g = 10 ** (drive_db / 20)
+ os = int(os or (8 if mode in ('hard', 'fold') else 4))
+ win = ('kaiser', 9.0)
+ u = signal.resample_poly(x2, os, 1, axis=0, window=win) if os > 1 else x2.copy()
+ if mode == 'tape':
+ u = _shelf_filter(u, 3000, 4.0, os)
+ b = 0.12 + bias
+ u = np.tanh(g * u + b) - np.tanh(b)
+ u = _shelf_filter(u, 3000, -4.0, os)
+ elif mode == 'hard':
+ u = np.clip(g * u + bias, -1, 1) - np.clip(bias, -1, 1)
+ elif mode == 'fold':
+ u = np.sin(np.pi / 2 * (g * u + bias)) - np.sin(np.pi / 2 * bias)
+ else:
+ u = np.tanh(g * u + bias) - np.tanh(bias)
+ y = signal.resample_poly(u, 1, os, axis=0, window=win)[:len(x2)] if os > 1 else u
+ if mode == 'tape':
+ y = butter(y, 'low', 17000, order=1)
+ if bias or mode == 'tape':
+ y = butter(y, 'high', 8, order=1)
+ if level == 'rms':
+ y *= _rms(x2) / _rms(y)
+ elif level == 'peak':
+ y *= np.abs(x2).max() / (np.abs(y).max() + 1e-12)
+ return _back(mix * y + (1 - mix) * x2, mono)
+
+
+def _shelf_filter(x, fc, gain_db, os=1):
+ b0, b1, a1 = _shelf(10 ** (gain_db / 20), fc / os, True)
+ return signal.lfilter([b0, b1], [1, a1], x, axis=0)
+
+
+@njit(cache=True)
+def _follow(x, att, rel):
+ y = np.empty_like(x)
+ e = 0.0
+ for i in range(len(x)):
+ a = att if x[i] > e else rel
+ e = a * e + (1 - a) * x[i]
+ y[i] = e
+ return y
+
+
+def _coef(ms):
+ return float(np.exp(-1.0 / max(1.0, ms * 1e-3 * SR)))
+
+
+def transient(x, attack_db=3.0, sustain_db=0.0, attack_ms=8.0, speed=1.0, floor_db=-60.0):
+ """transient shaper (level independent): attack_db boosts (or, negative, softens) the first ~attack_ms of every
+ onset, sustain_db lifts (or cuts) the decaying part progressively, e.g. a tail that dies too fast. Detection: the
+ envelope against a slow-rising copy of itself finds onsets, a slow- against a fast-release envelope finds decays
+ (a steady tone is left alone); speed scales the time constants; nothing below floor_db under the peak is lifted
+ (no pumped noise floor). Stereo is processed linked. Boosts can push peaks over 0 dBFS: limit afterwards.
+ y = fx.transient(impact, attack_db=4, sustain_db=5)"""
+ x2, mono = _2d(x)
+ m = np.ascontiguousarray(np.abs(x2).max(axis=1))
+ s = speed
+ ef = _follow(m, _coef(0.1 * s), _coef(15 * s))
+ es = _rise(ef, _coef(attack_ms * s))
+ a = np.clip((20 * np.log10((ef + 1e-9) / (es + 1e-9))) / 6.0, 0, 1)
+ el = _follow(m, _coef(0.3 * s), _coef(350 * s))
+ eq = _follow(m, _coef(0.3 * s), _coef(25 * s))
+ d = np.clip((20 * np.log10((el + 1e-9) / (eq + 1e-9))) / 9.0, 0, 1) * (1 - a)
+ gate = np.clip((20 * np.log10(el / (m.max() + 1e-12) + 1e-12) - floor_db) / 12.0, 0, 1)
+ gdb = attack_db * a + sustain_db * d * gate
+ gdb = _follow(gdb - gdb.min(), _coef(0.2), _coef(5)) + gdb.min() # de-zipper
+ return _back(x2 * 10 ** (gdb / 20)[:, None], mono)
+
+
+@njit(cache=True)
+def _rise(e, att):
+ """a slow-rising copy of envelope e that may never exceed it (follows every fall at once)"""
+ y = np.empty_like(e)
+ s = 0.0
+ for i in range(len(e)):
+ s = att * s + (1 - att) * e[i]
+ if s > e[i]:
+ s = e[i]
+ y[i] = s
+ return y
+
+
+@njit(cache=True)
+def _release(g, rel):
+ y = np.empty_like(g)
+ s = 1.0
+ for i in range(len(g)):
+ s = 1.0 - (1.0 - s) * rel
+ if g[i] < s:
+ s = g[i]
+ y[i] = s
+ return y
+
+
+def limit(x, ceiling_db=-1.0, lookahead=0.0025, release=0.08, true_peak=True):
+ """one-shot look-ahead peak limiter (offline, zero latency): the gain reaches the needed reduction smoothly
+ `lookahead` s before each peak and recovers with time constant `release`; true_peak measures 4x oversampled
+ peaks; a final safety clip keeps every sample at or under the ceiling. Stereo is linked.
+ y = fx.limit(mix, ceiling_db=-1.0)"""
+ x2, mono = _2d(x)
+ c = 10 ** (ceiling_db / 20)
+ pk = np.abs(x2).max(axis=1)
+ if true_peak:
+ up = np.abs(signal.resample_poly(x2, 4, 1, axis=0)).max(axis=1)
+ pk = np.maximum(pk, up[:len(up) // 4 * 4].reshape(-1, 4).max(axis=1)[:len(pk)])
+ need = np.minimum(1.0, c / np.maximum(pk, 1e-12))
+ L = max(1, n_of(lookahead))
+ g = ndimage.minimum_filter1d(need, 2 * L + 1, mode='nearest')
+ g = _release(np.ascontiguousarray(g), _coef(release * 1000))
+ g = ndimage.uniform_filter1d(g, 2 * L + 1, mode='nearest') # each average spans only gains <= the sample's need
+ y = x2 * g[:, None]
+ return _back(np.clip(y, -c, c), mono)
+
+
+def split(x, freqs):
+ """Linkwitz-Riley (4th order) band split at freqs (Hz, ascending): returns len(freqs) + 1 bands, low to high, that
+ sum back to an all-pass of x (flat magnitude), so bands can be processed and re-summed without a notch.
+ lo, mid, hi = fx.split(x, [200, 3000])"""
+ def lr4(v, f, kind):
+ sos = signal.butter(2, f, kind, fs=SR, output='sos')
+ return signal.sosfilt(sos, signal.sosfilt(sos, v, axis=0), axis=0)
+ bands, rest = [], np.asarray(x, np.float64)
+ for f in sorted(freqs):
+ bands = [lr4(b, f, 'low') + lr4(b, f, 'high') for b in bands] # same all-pass on the bands already split
+ bands.append(lr4(rest, f, 'low'))
+ rest = lr4(rest, f, 'high')
+ return bands + [rest]
+
+
+def _target_profile(target):
+ from .tone import ToneRef, profile
+ if isinstance(target, ToneRef):
+ return target.med
+ if isinstance(target, (list, tuple)) and len(target) and not np.isscalar(target[0]):
+ return ToneRef(list(target)).med
+ a = np.asarray(target, np.float64)
+ from .tone import CEN
+ return a if a.shape == (len(CEN),) else profile(a)
+
+
+def match_curve(x, target, amount=1.0, max_db=12.0, smooth=1):
+ """the per-band EQ gains (dB, tone.CEN 1/3-octave bands, 31.5 Hz - 16 kHz) that move x's tonal balance toward
+ `target`: a ToneRef, a list of reference files / arrays (their median profile), a 28-band profile (dB) or a
+ signal. Centred on the loud bands (level is not the point), smoothed over neighbours `smooth` times, clipped to
+ +-max_db; bands where x is near silent are not boosted more than 6 dB (EQ cannot create what is not there).
+ g = fx.match_curve(x, ['real_01.wav', 'real_02.wav'])"""
+ from .tone import profile
+ p, t = profile(x), _target_profile(target)
+ d = t - p
+ w = 10 ** (p / 10)
+ d -= np.sum(w * d) / np.sum(w)
+ d = np.where(p < -50, np.minimum(d, 6.0), d)
+ d = np.clip(d, -max_db, max_db)
+ for _ in range(int(smooth)):
+ d = np.convolve(np.pad(d, 1, mode='edge'), [0.25, 0.5, 0.25], 'valid')
+ return [round(float(v), 2) for v in amount * d]
+
+
+def eq_bands(x, gains_db, phase='min', taps=4096):
+ """apply per-band gains (dB at the tone.CEN centres, interpolated smoothly over log frequency, flat beyond the
+ ends) as one FIR: phase 'min' (minimum phase: no pre-ringing on transients, default) or 'linear'. Same length as x.
+ y = fx.eq_bands(x, fx.match_curve(x, refs))"""
+ from .tone import CEN
+ x = np.asarray(x, np.float64)
+ nf = taps * 4
+ f = np.fft.rfftfreq(nf, 1 / SR)
+ gdb = np.interp(np.log2(np.maximum(f, 1.0)), np.log2(CEN), np.asarray(gains_db, np.float64))
+ if phase == 'linear':
+ H = 10 ** (gdb / 20)
+ h = np.roll(np.fft.irfft(H, nf), taps // 2)[:taps] * np.hanning(taps)
+ delay = taps // 2
+ else:
+ c = np.fft.irfft(np.log(10 ** (gdb / 20)), nf) # real cepstrum -> fold -> minimum phase
+ fold = np.zeros(nf)
+ fold[0] = c[0]
+ fold[1:nf // 2] = 2 * c[1:nf // 2]
+ fold[nf // 2] = c[nf // 2]
+ h = np.fft.irfft(np.exp(np.fft.rfft(fold)), nf)[:taps]
+ h *= np.r_[np.ones(taps // 2), np.hanning(taps)[taps // 2:]]
+ delay = 0
+ y = signal.fftconvolve(x, h if x.ndim == 1 else h[:, None], axes=0)
+ return y[delay:delay + len(x)]
+
+
+def eq_match(x, target, amount=1.0, max_db=12.0, smooth=1, phase='min', iters=2):
+ """move x's 1/3-octave tonal balance toward `target` (see match_curve) with a smooth EQ; iterated (re-measured)
+ `iters` times; the output keeps the input's RMS.
+ y = fx.eq_match(made, ['kit/impact/real_01.wav', ...], amount=0.8)"""
+ from .tone import profile
+ x = np.asarray(x, np.float64)
+ tot = np.zeros(len(profile(x)))
+ y = x
+ for _ in range(max(1, int(iters))): # iterate the full correction, then scale it
+ tot = np.clip(tot + np.asarray(match_curve(y, target, 1.0, max_db, smooth)), -max_db, max_db)
+ y = eq_bands(x, tot, phase)
+ if amount != 1.0:
+ y = eq_bands(x, amount * tot, phase)
+ return y * (_rms(x) / _rms(y))
+
+
+# ------------------------------------------------------------------------------------------------ motion
+def _hermite(x, pos):
+ """cubic Hermite read of x (n,) or (n, c) at fractional positions (zeros outside)"""
+ x2, mono = _2d(x)
+ n = len(x2)
+ i = np.floor(pos).astype(np.int64)
+ fr = (pos - i)[:, None]
+ def at(k):
+ ok = (k >= 0) & (k < n)
+ v = np.zeros((len(k), x2.shape[1]))
+ v[ok] = x2[k[ok]]
+ return v
+ xm1, x0, x1, x2_ = at(i - 1), at(i), at(i + 1), at(i + 2)
+ c1 = 0.5 * (x1 - xm1)
+ c2 = xm1 - 2.5 * x0 + 2 * x1 - 0.5 * x2_
+ c3 = 0.5 * (x2_ - xm1) + 1.5 * (x0 - x1)
+ return _back(((c3 * fr + c2) * fr + c1) * fr + x0, mono)
+
+
+def doppler(x, speed=40.0, dist=4.0, t_close=None, air=1.0, width=0.8, stereo=True, c=343.0, floor_db=-30.0):
+ """a source passing the listener on a straight line: x is what the source emits (mono); speed m/s, closest
+ distance dist m, reached at source time t_close s (default the middle of x). Exact propagation delay (the pitch
+ falls as it passes), 1/distance level (0 dB at the closest point, never below floor_db), air absorption growing
+ with distance (high frequencies fade far away; air > 1 exaggerates it) and the pan following the angle (width).
+ Output starts when the first sound arrives.
+ car = fx.doppler(dsp.air(3, r, center=500), speed=30, dist=3)"""
+ xm = np.asarray(x, np.float64)
+ xm = xm.mean(axis=1) if xm.ndim == 2 else xm
+ T = len(xm) / SR
+ t0 = T / 2 if t_close is None else t_close
+ v = max(1e-3, min(speed, 0.95 * c))
+ rr = lambda te: np.hypot(v * (te - t0), dist)
+ t_first, t_last = rr(0.0) / c, T + rr(T) / c
+ t = t_first + np.arange(int((t_last - t_first) * SR)) / SR
+ b = t - t0
+ a = (c * c * b - np.sqrt(v * v * c * c * b * b + dist * dist * (c * c - v * v))) / (c * c - v * v)
+ te = a + t0
+ y = _hermite(xm, te * SR)
+ rad = rr(te)
+ y *= np.maximum(dist / rad, 10 ** (floor_db / 20))
+ att = air * 0.17 * np.maximum(rad - dist, 0) # dB lost at 10 kHz (air: about 0.17 dB / m)
+ fc = 10000 / np.maximum(10 ** (att / 10) - 1, 1e-9) ** 0.25
+ y = svf(y, np.clip(fc, 200, SR * 0.45), 0.707, 'low')
+ if not stereo:
+ return y
+ s = np.clip(v * (te - t0) / rad, -1, 1) * width
+ th = (s + 1) * np.pi / 4
+ return np.stack([y * np.cos(th), y * np.sin(th)], axis=1) * np.sqrt(2)
+
+
+def decorrelate(x, amount=1.0, cut=250.0):
+ """mono -> stereo: side = the signal through three short prime all-passes (high-passed at `cut`, the low end stays
+ mono), L = x + amount*side, R = x - amount*side. Deterministic.
+ st = fx.decorrelate(whoosh, 0.6)"""
+ xm = np.asarray(x, np.float64)
+ xm = xm.mean(axis=1) if xm.ndim == 2 else xm
+ s = np.ascontiguousarray(xm)
+ for d, g in ((149, 0.6), (263, -0.55), (397, 0.5)):
+ s = _allpass(s, d, g)
+ s = butter(s, 'high', cut)
+ s *= _rms(xm) / _rms(s) * 0.7
+ return np.stack([xm + amount * s, xm - amount * s], axis=1) / np.sqrt(1 + 0.49 * amount ** 2)
+
+
+def pan(x, pos):
+ """constant-power pan, static or moving: pos is -1..1 as a scalar, a per-sample array or breakpoints
+ [(t, pos), ...]; mono in -> stereo out; a stereo input is balanced.
+ st = fx.pan(whoosh, [(0, -0.9), (0.6, 0.9)])"""
+ x = np.asarray(x, np.float64)
+ p = np.clip(curve(pos, len(x)), -1, 1)
+ th = (p + 1) * np.pi / 4
+ g = np.stack([np.cos(th), np.sin(th)], axis=1) * np.sqrt(2)
+ return x[:, None] * g if x.ndim == 1 else x * g
+
+
+def width(x, w):
+ """stereo width over time (mid/side): w 0 = mono, 1 = as is, up to 2 = wider; scalar, array or breakpoints.
+ A mono input is decorrelated first.
+ st = fx.width(x, [(0, 0.2), (1.0, 1.4)])"""
+ x = np.asarray(x, np.float64)
+ if x.ndim == 1:
+ x = decorrelate(x)
+ k = np.clip(curve(w, len(x)), 0, 2)
+ m, s = (x[:, 0] + x[:, 1]) / 2, (x[:, 0] - x[:, 1]) / 2 * k
+ return np.stack([m + s, m - s], axis=1)
+
+
+def sweep_filter(x, f0, f1, q=4.0, curve_='exp', power=1.0, mode='bandn'):
+ """a resonant filter sweeping exponentially from f0 to f1 Hz over the length of x (riser: up, downlifter: down);
+ curve_ 'exp' | 'lin' | 'ease' with power (see dsp.sweep), mode as dsp.svf. Mono or stereo.
+ rise = fx.sweep_filter(dsp.noise(3, r, 'pink'), 300, 9000, q=6, curve_='ease', power=2)"""
+ from .dsp import sweep
+ x2, mono = _2d(x)
+ f = sweep(f0, f1, len(x2) / SR, curve_, power)[:len(x2)]
+ f = np.pad(f, (0, len(x2) - len(f)), mode='edge')
+ return _back(np.stack([svf(x2[:, c], f, q, mode) for c in range(x2.shape[1])], axis=1), mono)
+
+
+def varispeed(x, speed):
+ """play x at a speed that changes over time (1 = normal, 2 = an octave up, 0 = stopped): scalar, array or
+ breakpoints over the OUTPUT time; ends when x runs out or the speed array ends.
+ y = fx.varispeed(x, [(0, 1.0), (0.5, 0.25)])"""
+ x = np.asarray(x, np.float64)
+ if np.isscalar(speed):
+ n = int(len(x) / max(float(speed), 1e-3))
+ sp = np.full(n, float(speed))
+ else:
+ n = n_of(max(p[0] for p in speed)) if isinstance(speed, (list, tuple)) and np.ndim(speed[0]) else len(speed)
+ sp = np.maximum(curve(speed, n), 0)
+ pos = np.concatenate([[0.0], np.cumsum(sp[:-1])])
+ pos = pos[pos < len(x) - 1]
+ return _hermite(x, pos)
+
+
+def tape_stop(x, start=None, seconds=0.6, curve_=1.6):
+ """tape stop: normal speed until `start` s (default: so the stop ends with x), then the speed falls to 0 over
+ `seconds` (curve_ > 1 holds speed longer, then drops), the pitch sliding down; a tiny fade at the very end.
+ y = fx.tape_stop(chord, seconds=0.8)"""
+ x = np.asarray(x, np.float64)
+ T = len(x) / SR
+ start = max(0.0, T - seconds * 0.6) if start is None else start
+ n1, n2 = n_of(start), n_of(seconds)
+ u = np.linspace(0, 1, n2)
+ sp = np.concatenate([np.ones(n1), (1 - u) ** curve_])
+ pos = np.concatenate([[0.0], np.cumsum(sp[:-1])])
+ pos = pos[pos < len(x) - 1]
+ y = _hermite(x, pos)
+ k = min(len(y), n_of(0.005))
+ y[-k:] *= np.linspace(1, 0, k) if y.ndim == 1 else np.linspace(1, 0, k)[:, None]
+ return y
+
+
+# ------------------------------------------------------------------------------------------------ glitch + grains
+def crush(x, bits=8, rate=None, mix=1.0):
+ """bit-crush (quantise to `bits`, no dither: the grit is the point) and sample-rate reduction (sample-and-hold at
+ `rate` Hz, aliasing included on purpose).
+ y = fx.crush(x, bits=6, rate=8000)"""
+ x = np.asarray(x, np.float64)
+ y = x
+ if rate:
+ idx = np.floor(np.floor(np.arange(len(x)) * rate / SR) * SR / rate).astype(np.int64)
+ y = y[np.minimum(idx, len(x) - 1)]
+ if bits:
+ q = 2 ** (bits - 1)
+ y = np.round(y * q) / q
+ return mix * y + (1 - mix) * x
+
+
+def _fade_edges(seg, k):
+ k = min(k, len(seg) // 2)
+ if k > 1:
+ w = 0.5 - 0.5 * np.cos(np.linspace(0, np.pi, k))
+ w = w if seg.ndim == 1 else w[:, None]
+ seg[:k] *= w
+ seg[-k:] *= w[::-1]
+ return seg
+
+
+def stutter(x, start, length, repeats=4, gate=0.85, decay_db=0.0, pitch_step=0.0, fade=0.002):
+ """stutter: the slice [start, start + length) s plays `repeats` times (each gated to `gate` of its slot, decay_db
+ quieter and pitch_step semitones higher than the one before; micro-fades, no clicks), then x continues.
+ y = fx.stutter(x, 0.0, 0.06, repeats=6, decay_db=-2, pitch_step=1)"""
+ x = np.asarray(x, np.float64)
+ i0, L = n_of(start) if start > 0 else 0, n_of(length)
+ sl = x[i0:i0 + L]
+ parts = [x[:i0]]
+ for k in range(int(repeats)):
+ rate = 2 ** (k * pitch_step / 12)
+ seg = _hermite(sl, np.arange(L) * rate) if pitch_step else sl.copy()
+ g = max(2, int(L * gate))
+ seg[g:] = 0
+ seg[:g] = _fade_edges(seg[:g].copy(), n_of(fade))
+ parts.append(seg * 10 ** (k * decay_db / 20))
+ parts.append(x[i0 + L:])
+ return np.concatenate(parts)
+
+
+def glitch(x, r, grid=0.03, repeat=0.25, reverse=0.1, drop=0.08, octave=0.08, crush_=0.08, fade=0.0015):
+ """slice x on a `grid` (s) and re-arrange it: each slice is, with these probabilities, a repeat of the previous
+ output slice, reversed, dropped (silence), an octave up (played twice at double speed) or crushed (6 bit, 6 kHz);
+ otherwise kept. Micro-fades on every edge. Same length as x; deterministic for a given rng.
+ g = fx.glitch(chord, dsp.rng(3), grid=0.045, repeat=0.3)"""
+ x = np.asarray(x, np.float64)
+ L = max(16, n_of(grid))
+ out = np.zeros_like(x)
+ prev = None
+ k = n_of(fade)
+ for i in range(0, len(x), L):
+ seg = x[i:i + L].copy()
+ u = r.uniform()
+ acts = np.cumsum([repeat, reverse, drop, octave, crush_])
+ if u < acts[0] and prev is not None and len(prev) == len(seg):
+ seg = prev.copy()
+ elif u < acts[1]:
+ seg = seg[::-1].copy()
+ elif u < acts[2]:
+ seg[:] = 0
+ elif u < acts[3]:
+ half = seg[::2]
+ seg = np.concatenate([_fade_edges(half.copy(), k), _fade_edges(half.copy(), k)])[:len(seg)]
+ seg = np.pad(seg, [(0, len(x[i:i + L]) - len(seg))] + [(0, 0)] * (seg.ndim - 1))
+ elif u < acts[4]:
+ seg = crush(seg, 6, 6000)
+ seg = _fade_edges(seg, k)
+ out[i:i + len(seg)] = seg
+ prev = seg
+ return out
+
+
+def granular(src, seconds, r, grain=0.06, density=40.0, pos=(0.0, 1.0), pos_jitter=0.02, pitch=0.0, pitch_jitter=0.0,
+ time_jitter=0.5, spread=0.6, reverse=0.0):
+ """granular cloud over a computed source: `density` grains per second of `grain` s (Hann windows), read from
+ position `pos` (fraction of src: scalar, (start, end) scanned over the output, or breakpoints) +- pos_jitter,
+ transposed pitch +- pitch_jitter semitones, onsets jittered by time_jitter (0 = a regular train, 1 = random),
+ panned randomly within +-spread, a `reverse` share played backwards. Stereo, RMS about the source's.
+ cloud = fx.granular(dsp.bell(880, 1, r), 3.0, r, grain=0.08, density=60, pos=(0.05, 0.6), pitch_jitter=0.3)"""
+ s = np.asarray(src, np.float64)
+ s = s.mean(axis=1) if s.ndim == 2 else s
+ n = n_of(seconds)
+ out = np.zeros((n, 2))
+ if isinstance(pos, tuple) and len(pos) == 2 and np.isscalar(pos[0]):
+ pc = np.linspace(pos[0], pos[1], n)
+ else:
+ pc = curve(pos, n)
+ period = 1.0 / max(density, 1e-3)
+ t = 0.0
+ gl = n_of(grain)
+ win = np.hanning(gl)
+ count = 0
+ while t < seconds:
+ tt = t + time_jitter * period * r.uniform(-0.5, 0.5)
+ i = n_of(max(tt, 0.0)) if tt > 0 else 0
+ if i < n:
+ rate = 2 ** ((pitch + pitch_jitter * r.standard_normal()) / 12)
+ p0 = (pc[min(i, n - 1)] + pos_jitter * r.standard_normal()) * (len(s) - gl * rate)
+ idx = np.clip(p0, 0, max(0, len(s) - gl * rate - 2)) + np.arange(gl) * rate
+ gr = _hermite(s, idx) * win
+ if r.uniform() < reverse:
+ gr = gr[::-1]
+ th = (np.clip(spread * r.uniform(-1, 1), -1, 1) + 1) * np.pi / 4
+ m = min(gl, n - i)
+ out[i:i + m, 0] += gr[:m] * np.cos(th) * np.sqrt(2)
+ out[i:i + m, 1] += gr[:m] * np.sin(th) * np.sqrt(2)
+ count += 1
+ t += period
+ overlap = max(1.0, density * grain * 0.375) # Hann power is 3/8 of the rectangle's
+ return out / np.sqrt(overlap)
+
+
+def reverse_swell(hit, r, seconds=1.0, rt60=2.5, size=0.8, damp=0.4, swell_db=-6.0, gap=0.0, hit_room=0.15):
+ """a reverse swell into a hit: the hit's own computed reverb tail, reversed, rises for `seconds` and lands on the
+ hit, whose onset sits exactly at `seconds` + gap in the output; swell_db sets the swell's peak against the hit's,
+ hit_room adds a little of the same space after the hit. Stereo.
+ y = fx.reverse_swell(dsp.sub_drop(1.2, r), r, seconds=1.2)"""
+ h = np.asarray(hit, np.float64)
+ hm = h.mean(axis=1) if h.ndim == 2 else h
+ wet = reverb(hm, r, rt60=rt60, size=size, damp=damp, predelay=0.0, wet=1.0, dry=0.0, stereo=True,
+ tail=max(seconds, rt60) + 0.1)
+ k = n_of(seconds)
+ sw = wet[:k][::-1].copy()
+ f = min(k, n_of(seconds * 0.35))
+ sw[:f] *= (np.linspace(0, 1, f) ** 2)[:, None]
+ hs = np.stack([h, h], axis=1) if h.ndim == 1 else h
+ after = reverb(hm, r, rt60=rt60 * 0.6, size=size, damp=damp, predelay=0.0, wet=hit_room, dry=1.0,
+ stereo=True) if hit_room else hs
+ sw *= np.abs(after).max() * 10 ** (swell_db / 20) / (np.abs(sw).max() + 1e-12)
+ out = np.zeros((k + n_of(gap) * (gap > 0) + len(after), 2))
+ place(out, sw, 0.0)
+ place(out, after, seconds + gap)
+ return out
diff --git a/earshot/forge/polish.py b/earshot/forge/polish.py
new file mode 100644
index 0000000..9235f0b
--- /dev/null
+++ b/earshot/forge/polish.py
@@ -0,0 +1,577 @@
+"""Artifact detectors and polish measures for sound effects, measured straight from the samples (no model).
+
+ from earshot.forge.polish import PolishRef, measure, loudness_spread
+ m = measure(x) # every detector's raw number for one 48 kHz sound (mono or stereo)
+ pr = PolishRef(['ref_01.wav', ...]) # the refs' own numbers loosen the tolerances (PolishRef([]): absolute)
+ c = pr.check(x)
+ c['art_pen'], c['polish_pen'], c['hard'], c['words'] # e.g. 'artifacts: 2 clicks (0.212 s, 0.305 s); polish:
+ # attack 41 ms (refs 1-9), crest 19.5 dB (refs 11-15)'
+ loudness_spread([x1, x2, x3]) # do the variants of one sound play at one loudness?
+
+Why: a forged sound effect is judged in a motion-design edit next to pro library sounds. A click from a hard gate, a
+foldover whistle on a sweep, a 3 kHz ring, a tail chopped at -30 dB or a spectrum frozen while the refs move are heard
+at once and none of earshot's model checks reliably hears them (they were trained to tell real from synthetic, not
+good from broken). Each detector below is a separate function returning plain data.
+
+Artifacts (`measure` keys):
+ clicks isolated discontinuities inside the sound: 1 ms frames of the 2nd difference (+12 dB/oct, so a jump
+ stands far above smooth content) more than 18 dB above both neighbourhoods while the neighbourhoods
+ agree within 8 dB, with 60 % or more of that frame's energy in 2 samples (a jump, not a physical tick,
+ which spreads over a fraction of a ms); an onset, where the sound starts and stays, is not a click and
+ the loudest transient +-3 ms is the sound's own attack. edge_db: the first / last sample re the peak (a file
+ that starts or ends off zero clicks when played)
+ fold aliasing / foldover: in clearly harmonic frames (4096-point STFT; >= 70 % of the 60 Hz - 4 kHz energy
+ within +-2 bins of one harmonic grid, f0 150 Hz - 3 kHz, refined by least squares on the low partials),
+ the share of the 4-20 kHz energy that lies off the grid. Each harmonic's window is widened by its sweep
+ smear (from the f0 change between frames); harmonics whose window would exceed half the spacing are not
+ judged. A naive (non band-limited) oscillator folds its upper partials back between the harmonics.
+ Noise, modal and inharmonic sounds have no grid and are not judged.
+ nyq_db: energy share above 20 kHz (recordings roll off; naive oscillators and white noise do not)
+ dc_db |mean| re the peak; infra_db: energy share below 15 Hz (a one-sided puff, an offset, drift)
+ res_db the harshest narrow resonance: energy spectrum (4096-point frames) minus its median over +-1/3 octave,
+ weighted by where the ear is touchy (1 for 2-6 kHz, 0.7 for 1-2 and 6-9 kHz, 0.5 elsewhere), peaks
+ within 30 dB of the spectrum's maximum only; res_hz its frequency
+ evolution how much the spectral SHAPE changes: per 10 ms frame (30 ms smoothed) the octave-group profile re the
+ frame's own total (so a fade does not count), change over 50 ms as an RMS over groups weighted by their
+ energy share (quiet groups' estimation noise does not count), energy-weighted over active frames (dB);
+ centroid_oct: the 10-90 % range of the spectral centroid in octaves. Frozen = below 0.35 x the refs'
+ median where the refs move (median >= 1.5 dB); penalty 2 x the relative shortfall
+ ring_z comb-like metallic ringing in the tail (from 30 ms after the peak to -50 dB): cepstral peak of the
+ tail's log spectrum (300 Hz - 12 kHz, trend removed) at 2-40 ms quefrency, as a robust z against the
+ rest of that range; ring_ms its delay. A dense natural tail has no repeating delay
+ trunc_db a truncated tail: the level re the peak at which the decay was cut (the last 10 ms if the sound was
+ still decaying there, or the level before a final fade that falls much faster than the decay before
+ it); None when the sound reaches -55 dB or its own floor
+ silence_ms digital silence inside the sound (runs of |x| < 1e-6 of >= 2 ms between the first and last audible
+ sample), total ms; a recording always has a floor, a gap of exact zeros is an edit or a gate
+ clip_runs runs of >= 3 equal samples at the peak (flat tops), per channel; overs: samples beyond 1.0
+Polish:
+ attack_ms 10 -> 90 % rise of the 0.5 ms RMS envelope to its first near-peak (sample-accurate crispness)
+ onset_hf_db energy share above 6 kHz in the first 10 ms from the onset (the snap / top end of the attack)
+ bump_db the largest re-rise of the 10 ms envelope above its running minimum after the peak (frames above
+ -45 dB): a smooth tail falls monotonically; decay_db_s the fitted decay rate, t40_ms the tail length
+ crest_db peak - the loudest 50 ms of K-weighted RMS (high = thin, spiky, little body at the same peak)
+ loudness_spread(xs): momentary-max LUFS (K-weighted, 400 ms) per variant at its own level, spread and std (LU)
+
+Penalties (`PolishRef.check`): RULES give each measure a bad direction, an absolute tolerance, a margin, a scale and a
+cap. limit = the absolute tolerance, loosened to the refs' worst value + margin when refs are given (a trait the real
+recordings share is not a defect), or for the relative-only measures (attack, onset_hf, t40, crest, evolution) set by
+the refs alone (dB margins widen to a quarter of the refs' own range: 3-6 diverse refs pin a range down loosely);
+penalty = min(cap, excess / scale) (log2 ratios for times). 'hard' artifacts at penalty >= 0.5 are listed
+in check()['hard'] and fail the critic's `passes`; designed=True (glitch, stutter) makes clicks and digital silence
+soft at half weight. art_pen = sum of artifact penalties; polish_pen = sum of polish penalties.
+"""
+import numpy as np
+
+SR = 48000
+K_SOS = np.array([[1.53512485958697, -2.69169618940638, 1.19839281085285, 1.0, -1.69065929318241, 0.73248077421585],
+ [1.0, -2.0, 1.0, 1.0, -1.99004745483398, 0.99007225036621]]) # ITU-R BS.1770 K-weighting, 48 kHz
+
+
+def _mono(x):
+ x = np.asarray(x, np.float64)
+ return x.mean(axis=1) if x.ndim == 2 else x
+
+
+def _chans(x):
+ x = np.asarray(x, np.float64)
+ return x.T if x.ndim == 2 else x[None]
+
+
+def _load(x):
+ if isinstance(x, str):
+ from .. import audio
+ return audio.load(x)
+ return x
+
+
+def _fdb(m, n):
+ """dB power of consecutive n-sample frames"""
+ k = len(m) // n
+ if k < 1:
+ return np.full(1, -240.0)
+ return 10 * np.log10((m[:k * n].reshape(k, n) ** 2).mean(1) + 1e-24)
+
+
+def _stft(m, nfft, hop):
+ if len(m) < nfft:
+ m = np.pad(m, (0, nfft - len(m)))
+ k = 1 + (len(m) - nfft) // hop
+ idx = np.arange(nfft)[None, :] + hop * np.arange(k)[:, None]
+ return np.abs(np.fft.rfft(m[idx] * np.hanning(nfft), axis=1)) ** 2
+
+
+# ------------------------------------------------------------------------------------------------ artifacts
+def clicks(x, thr=18.0):
+ """isolated discontinuities inside the sound (count, times) and the edge jump (dB re peak)"""
+ m = _mono(x)
+ peak = np.abs(m).max() + 1e-12
+ edge = 20 * np.log10(max(abs(m[0]), abs(m[-1])) / peak + 1e-12)
+ d2 = np.diff(m, 2)
+ e = _fdb(d2, 48)
+ lvl = _fdb(m, 48)
+ k = min(len(e), len(lvl))
+ times = []
+ if k >= 24:
+ e, lvl = e[:k], lvl[:k]
+ ip = int(np.argmax(lvl))
+ P = np.pad(e, 12, mode='edge')
+ W = np.lib.stride_tricks.sliding_window_view(P, 25) # W[i] = e[i-12 .. i+12]
+ pre, post = np.median(W[:, 2:10], 1), np.median(W[:, 15:23], 1)
+ q = np.sort(d2[:k * 48].reshape(k, 48) ** 2, axis=1)
+ imp = q[:, -2:].sum(1) / (q.sum(1) + 1e-30) # a jump puts its energy in 2 samples
+ hit = ((e - np.maximum(pre, post) > thr) & (np.abs(post - pre) < 8) & (lvl > lvl.max() - 70)
+ & (np.abs(np.arange(k) - ip) > 3) & (imp > 0.6))
+ i = 0
+ while i < k:
+ if hit[i]:
+ j = i
+ while j + 1 < k and hit[j + 1]:
+ j += 1
+ times.append(round((i + j + 1) / 2 / 1000, 3))
+ i = j + 1
+ else:
+ i += 1
+ return {'clicks': len(times), 'click_times': times[:8], 'edge_db': round(float(edge), 1)}
+
+
+def _f0(P, Pw, f, c0, low, G):
+ """harmonic grid of one spectrum frame: (f0 refined by least squares on the low partials, grid share) or None"""
+ tl = P[(f >= 60) & (f < 4000)].sum()
+ if tl <= 0:
+ return None
+ on = np.where(low, Pw[G], 0.0).sum(1) / tl - 0.02 * low.sum(1) # a dense grid covers more by chance
+ i = int(np.argmax(on))
+ if on[i] < 0.7:
+ return None
+ df = f[1]
+ hs, fs, ws = [], [], []
+ for h in range(1, int(4000 / c0[i]) + 1):
+ k = int(round(h * c0[i] / df))
+ lo, hi = max(1, k - 3), min(len(P) - 2, k + 3)
+ j = lo + int(np.argmax(P[lo:hi + 1]))
+ a, b, c = np.log(P[j - 1:j + 2] + 1e-30)
+ hs.append(h)
+ fs.append((j + 0.5 * (a - c) / (a - 2 * b + c - 1e-12)) * df)
+ ws.append(P[j])
+ hs, fs, ws = np.array(hs, float), np.array(fs), np.array(ws)
+ return float((ws * hs * fs).sum() / (ws * hs * hs).sum()), float(on[i])
+
+
+def aliasing(x, nfft=4096, hop=1024, frames=64):
+ """fold: in clearly harmonic frames, the share of the 4-20 kHz energy that lies OFF the harmonic grid (each
+ harmonic's window widened by its sweep smear); nyq_db: energy share above 20 kHz"""
+ m = _mono(x)
+ S = _stft(m, nfft, hop)
+ f = np.fft.rfftfreq(nfft, 1 / SR)
+ df = f[1]
+ nyq = 10 * np.log10(S[:, f >= 20000].sum() / (S.sum() + 1e-30) + 1e-12)
+ lvl = 10 * np.log10(S.sum(1) + 1e-30)
+ act = np.nonzero(lvl > lvl.max() - 30)[0][:frames]
+ Pw = sum(np.roll(S, k, axis=1) for k in (-2, -1, 0, 1, 2)) # energy within +-2 bins
+ c0 = 150 * 2 ** (np.arange(0, np.log2(3000 / 150), 1 / 96))
+ H = np.arange(1, int(4000 / c0[0]) + 1)
+ hf = c0[:, None] * H[None, :]
+ G = np.clip(np.rint(hf / df).astype(int), 0, len(f) - 1)
+ low = (hf < 4000) & (hf >= 60)
+ f0 = {int(t): _f0(S[t], Pw[t], f, c0, low, G) for t in act}
+ C = np.cumsum(S, axis=1)
+ num = den = 0.0
+ harm = 0
+ for t, r in f0.items():
+ if r is None:
+ continue
+ nb = [f0.get(t + d) for d in (-1, 1)]
+ rates = [abs(q[0] - r[0]) / (hop / SR) for q in nb if q is not None]
+ rate = max(rates) if rates else 0.0 # Hz / s of the fundamental
+ hs = np.arange(int(np.ceil(4000 / r[0])), int(20000 / r[0]) + 1)
+ hw = 2 + 0.5 * hs * rate * (nfft / SR) / df # half-width in bins, with smear
+ ok = 2 * hw * df < 0.5 * r[0]
+ if not ok.any() or not ok[0]:
+ continue
+ hs, hw = hs[ok], hw[ok]
+ if len(hs) > 1 and (np.diff(hs) > 1).any():
+ cut = int(np.nonzero(np.diff(hs) > 1)[0][0]) + 1
+ hs, hw = hs[:cut], hw[:cut]
+ top = min(20000.0, (hs[-1] + 0.5) * r[0])
+ b0, b1 = int(np.searchsorted(f, 4000)), int(np.searchsorted(f, top))
+ th = C[t, b1] - C[t, b0]
+ if b1 - b0 < 20 or th < 1e-3 * C[t, -1]:
+ continue
+ k = hs * r[0] / df
+ lo = np.clip(np.floor(k - hw).astype(int), b0, b1)
+ hi = np.clip(np.ceil(k + hw).astype(int), b0, b1)
+ on = float((C[t, hi] - C[t, lo]).sum())
+ harm += 1
+ num += (1.0 - min(1.0, on / th)) * th
+ den += th
+ fold = num / den if den > 0 and harm >= 2 else 0.0
+ return {'fold': round(float(fold), 3), 'harmonic_frames': harm, 'nyq_db': round(float(nyq), 1)}
+
+
+def dc(x):
+ m = _mono(x)
+ peak = np.abs(m).max() + 1e-12
+ from scipy.fft import next_fast_len
+ n = next_fast_len(len(m), real=True)
+ F = np.abs(np.fft.rfft(m, n)) ** 2
+ f = np.fft.rfftfreq(n, 1 / SR)
+ return {'dc_db': round(float(20 * np.log10(abs(m.mean()) / peak + 1e-12)), 1),
+ 'infra_db': round(float(10 * np.log10(F[f < 15].sum() / (F.sum() + 1e-30) + 1e-12)), 1)}
+
+
+def resonances(x, top=3):
+ """the harshest narrow peak above the 2/3-octave median of the energy spectrum (weighted dB, Hz)"""
+ m = _mono(x)
+ nfft = 4096 if len(m) >= 4096 else 1 << max(10, int(np.ceil(np.log2(max(len(m), 2)))))
+ S = _stft(m, nfft, nfft // 4).sum(0)
+ f = np.fft.rfftfreq(nfft, 1 / SR)
+ L = 10 * np.log10(S + 1e-30)
+ g = 100 * 2 ** (np.arange(0, np.log2(16000 / 100), 1 / 48)) # 48 points per octave, 100 Hz - 16 kHz
+ edges = np.searchsorted(f, np.concatenate([g / 2 ** (1 / 96), [g[-1] * 2 ** (1 / 96)]]))
+ Lg = np.array([L[a:max(b, a + 1)].max() for a, b in zip(edges[:-1], edges[1:])])
+ from scipy.ndimage import median_filter
+ prom = Lg - median_filter(Lg, size=33, mode='nearest')
+ w = np.where((g >= 2000) & (g <= 6000), 1.0, np.where((g >= 1000) & (g <= 9000), 0.7, 0.5))
+ ok = (Lg > Lg.max() - 30) & (prom > 0)
+ loc = np.r_[False, (prom[1:-1] >= prom[:-2]) & (prom[1:-1] >= prom[2:]), False] & ok
+ idx = np.nonzero(loc)[0]
+ if not len(idx):
+ return {'res_db': 0.0, 'res_hz': None, 'res_peaks': []}
+ order = idx[np.argsort(-(prom[idx] * w[idx]))][:top]
+ return {'res_db': round(float(prom[order[0]] * w[order[0]]), 1), 'res_hz': round(float(g[order[0]])),
+ 'res_peaks': [(round(float(g[i])), round(float(prom[i]), 1)) for i in order]}
+
+
+def evolution(x, G=None):
+ """how much the spectral shape changes over the sound (dB), and the centroid's range (octaves)"""
+ from .tone import CEN, bandgram
+ G = bandgram(x) if G is None else G
+ P = 10 ** (G / 10)
+ groups = [P[:, (CEN >= a) & (CEN < b)].sum(1) for a, b in
+ ((0, 112), (112, 224), (224, 447), (447, 894), (894, 1789), (1789, 3578), (3578, 7155), (7155, 1e5))]
+ Q = np.stack(groups, 1) + 1e-12
+ Q = np.apply_along_axis(lambda v: np.convolve(v, np.ones(3) / 3, mode='same'), 0, Q) # 30 ms smoothing
+ tot = Q.sum(1)
+ act = 10 * np.log10(tot + 1e-30) > 10 * np.log10(tot.max()) - 30
+ if act.sum() < 8:
+ return {'evolution_db': None, 'centroid_oct': None}
+ R = 10 * np.log10(Q / tot[:, None])
+ R = np.maximum(R, -40)
+ lag = 5
+ i = np.nonzero(act)[0]
+ i = i[(i >= lag) & act[np.clip(i - lag, 0, None)]]
+ if len(i) < 3:
+ return {'evolution_db': None, 'centroid_oct': None}
+ sh = 0.5 * (Q[i] / tot[i, None] + Q[i - lag] / tot[i - lag, None]) # groups weighted by their share
+ ch = np.sqrt((sh * (R[i] - R[i - lag]) ** 2).sum(1) / sh.sum(1))
+ w = tot[i]
+ cen = np.log2((P[act] * CEN).sum(1) / P[act].sum(1))
+ lo, hi = np.percentile(cen, [10, 90])
+ return {'evolution_db': round(float((ch * w).sum() / w.sum()), 2), 'centroid_oct': round(float(hi - lo), 2)}
+
+
+def ring(x):
+ """comb-like periodicity of the tail: cepstral peak z at 2-40 ms quefrency (None when the tail is too short)"""
+ m = _mono(x)
+ e = _fdb(m, 480)
+ ip = int(np.argmax(e))
+ s = ip + 3
+ below = np.nonzero(e[s:] < e[ip] - 50)[0]
+ end = s + (int(below[0]) if len(below) else len(e) - s)
+ if end - s < 6:
+ return {'ring_z': None, 'ring_ms': None}
+ seg = m[s * 480:end * 480][:16384]
+ N = 8192 if len(seg) <= 8192 else 16384
+ L = 10 * np.log10(np.abs(np.fft.rfft(seg * np.hanning(len(seg)), N)) ** 2 + 1e-30)
+ f = np.fft.rfftfreq(N, 1 / SR)
+ sel = (f >= 300) & (f <= 12000)
+ Lc = np.maximum(L[sel], L[sel].max() - 60)
+ df = f[1]
+ from scipy.ndimage import uniform_filter1d
+ r = Lc - uniform_filter1d(Lc, max(3, int(2000 / df)), mode='nearest')
+ C = np.abs(np.fft.rfft(r * np.hanning(len(r))))
+ q = np.arange(len(C)) / (len(r) * df)
+ band = (q >= 0.002) & (q <= 0.04)
+ if band.sum() < 8:
+ return {'ring_z': None, 'ring_ms': None}
+ c = C[band]
+ mad = np.median(np.abs(c - np.median(c))) * 1.4826 + 1e-12
+ j = int(np.argmax(c))
+ return {'ring_z': round(float((c[j] - np.median(c)) / mad), 1), 'ring_ms': round(float(q[band][j] * 1000), 2)}
+
+
+def truncated(x):
+ """trunc_db: the level (dB re peak) at which a still-decaying tail was cut, or None"""
+ m = _mono(x)
+ e = _fdb(m, 240) # 5 ms frames
+ if len(e) < 12:
+ return {'trunc_db': None, 'end_db': None}
+ ip = int(np.argmax(e))
+ e = e - e[ip]
+ end = float(np.mean(e[-2:]))
+ if len(e) - ip < 12 or end < -55:
+ return {'trunc_db': None, 'end_db': round(end, 1)}
+ before = e[max(ip, len(e) - 38):len(e) - 8] # 150 ms .. 40 ms before the end
+ slope = np.polyfit(np.arange(len(before)) * 0.005, before, 1)[0] if len(before) >= 4 else 0.0 # dB/s
+ fade_from = float(np.mean(e[-10:-7]))
+ expect = max(3.0, -slope * 0.04) # natural drop over the last 40 ms
+ cut = None
+ if fade_from - end > 3 * expect + 6 and fade_from > -50: # a final fade much faster than the decay
+ cut = fade_from
+ elif slope < -20 and end > -50: # still decaying when the file stops
+ cut = end
+ return {'trunc_db': None if cut is None else round(cut, 1), 'end_db': round(end, 1)}
+
+
+def silence(x):
+ m = _mono(x)
+ a = np.abs(m)
+ peak = a.max() + 1e-12
+ aud = np.nonzero(a > peak * 1e-3)[0]
+ if len(aud) < 2:
+ return {'silence_ms': 0.0, 'silence_max_ms': 0.0}
+ z = a[aud[0]:aud[-1]] < 1e-6
+ runs = np.diff(np.r_[0, z.astype(np.int8), 0])
+ st, en = np.nonzero(runs == 1)[0], np.nonzero(runs == -1)[0]
+ L = (en - st)[(en - st) >= 96]
+ return {'silence_ms': round(float(L.sum() / 48), 1),
+ 'silence_max_ms': round(float(L.max() / 48), 1) if len(L) else 0.0}
+
+
+def clipping(x):
+ runs, overs = 0, 0
+ for c in _chans(x):
+ a = np.abs(c)
+ pk = a.max()
+ if pk <= 0:
+ continue
+ overs += int((a > 1.0).sum())
+ flat = (a[1:] >= 0.99 * pk) & (np.abs(np.diff(c)) < 2e-6)
+ d = np.diff(np.r_[0, flat.astype(np.int8), 0])
+ st, en = np.nonzero(d == 1)[0], np.nonzero(d == -1)[0]
+ runs += int(((en - st) >= 2).sum())
+ return {'clip_runs': runs, 'overs': overs}
+
+
+# ------------------------------------------------------------------------------------------------ polish
+def attack(x):
+ """10 -> 90 % rise (ms) to the first near-peak of the 0.5 ms RMS envelope, and the HF share of the first 10 ms"""
+ m = _mono(x)
+ n = len(m) // 12
+ if n < 4:
+ return {'attack_ms': None, 'onset_hf_db': None}
+ p = (m[:n * 12].reshape(n, 12) ** 2).mean(1) # 0.25 ms hops
+ env = np.sqrt(np.convolve(p, np.ones(2) / 2, mode='same')) # 0.5 ms RMS
+ top = env.max()
+ i90 = int(np.nonzero(env >= 0.9 * top)[0][0])
+ pre = np.nonzero(env[:i90 + 1] < 0.1 * top)[0]
+ i10 = int(pre[-1]) if len(pre) else 0
+ on = i10 * 12
+ seg = m[on:on + 480]
+ F = np.abs(np.fft.rfft(seg * np.hanning(len(seg)), 1024)) ** 2 if len(seg) >= 32 else None
+ hf = None
+ if F is not None:
+ f = np.fft.rfftfreq(1024, 1 / SR)
+ hf = round(float(10 * np.log10(F[f >= 6000].sum() / (F.sum() + 1e-30) + 1e-12)), 1)
+ return {'attack_ms': round((i90 - i10) * 0.25, 2), 'onset_hf_db': hf}
+
+
+def tail(x):
+ """bump_db (re-rise above the running minimum after the peak), decay_db_s, t40_ms"""
+ m = _mono(x)
+ e = _fdb(m, 480)
+ if len(e) < 4:
+ return {'bump_db': 0.0, 'decay_db_s': None, 't40_ms': None}
+ p = np.convolve(10 ** (e / 10), np.ones(3) / 3, mode='same')
+ e = 10 * np.log10(p + 1e-24)
+ ip = int(np.argmax(e))
+ e = e - e[ip]
+ after = e[ip:]
+ below = np.nonzero(after < -40)[0]
+ t40 = float(below[0] * 10) if len(below) else float(len(after) * 10)
+ run = np.minimum.accumulate(after)
+ live = after > -45
+ live[:2] = False
+ bump = float(np.max((after - run)[live])) if live.any() else 0.0
+ sel = np.nonzero((after <= -6) & (after >= -45))[0]
+ rate = None
+ if len(sel) >= 3:
+ rate = round(float(np.polyfit(sel * 0.01, after[sel], 1)[0]), 1)
+ return {'bump_db': round(bump, 1), 'decay_db_s': rate, 't40_ms': t40}
+
+
+def _movmax(p, n):
+ """the largest mean of n consecutive values (the whole mean when shorter)"""
+ if len(p) < n:
+ return float(p.sum() / n)
+ c = np.concatenate([[0.0], np.cumsum(p)])
+ return float((c[n:] - c[:-n]).max() / n)
+
+
+def _kweight(m):
+ from scipy import signal
+ return signal.sosfilt(K_SOS, m)
+
+
+def body(x):
+ """crest_db: peak (dBFS) minus the loudest 50 ms of K-weighted RMS"""
+ m = _mono(x)
+ k = _kweight(m)
+ loud = 10 * np.log10(_movmax(k ** 2, 2400) + 1e-24)
+ return {'crest_db': round(float(20 * np.log10(np.abs(m).max() + 1e-12) - loud), 1)}
+
+
+def lufs_max(x):
+ """momentary-max loudness (LUFS, K-weighted 400 ms windows over the channel sum of powers; zero-padded when
+ the sound is shorter than 400 ms, as a meter would show it)"""
+ tot = sum(_kweight(c) ** 2 for c in _chans(x))
+ return float(-0.691 + 10 * np.log10(_movmax(tot, 19200) + 1e-24))
+
+
+def loudness_spread(xs):
+ """do variants of one sound play at one loudness? per-variant momentary-max LUFS at their own levels"""
+ L = [lufs_max(_load(x)) for x in xs]
+ if not L:
+ return {'lufs': [], 'spread_lu': None, 'std_lu': None}
+ return {'lufs': [round(v, 2) for v in L], 'spread_lu': round(float(max(L) - min(L)), 2),
+ 'std_lu': round(float(np.std(L)), 2)}
+
+
+def measure(x, G=None):
+ """every detector's raw numbers for one sound (G: its tone.bandgram, when already computed)"""
+ x = _load(x)
+ out = {}
+ for fn in (clicks, aliasing, dc, resonances, ring, truncated, silence, clipping, attack, tail, body):
+ out.update(fn(x))
+ out.update(evolution(x, G))
+ return out
+
+
+# ------------------------------------------------------------------------------------------------ penalties
+# key bad abs margin scale kind cap (bad: +1 high is bad, -1 low is bad)
+RULES = (('clicks', +1, 0.0, 0.0, 1.0, 'hard', 2.0),
+ ('edge_db', +1, -40.0, 6.0, 10.0, 'hard', 2.0),
+ ('fold', +1, 0.02, 0.02, 0.03, 'hard', 2.0),
+ ('nyq_db', +1, -35.0, 6.0, 8.0, 'soft', 1.0), # inaudible itself: a sign of naive synthesis
+ ('dc_db', +1, -40.0, 6.0, 10.0, 'hard', 2.0),
+ ('infra_db', +1, -20.0, 6.0, 8.0, 'soft', 1.5), # wasted headroom, thumps on real systems
+ ('res_db', +1, 15.0, 3.0, 5.0, 'soft', 2.0),
+ ('ring_z', +1, 12.0, 3.0, 6.0, 'soft', 2.0),
+ ('trunc_db', +1, -45.0, 6.0, 10.0, 'hard', 2.0),
+ ('silence_ms', +1, 2.0, 2.0, 10.0, 'hard', 2.0),
+ ('clip_runs', +1, 0.0, 1.0, 4.0, 'hard', 2.0),
+ ('evolution_db', -1, None, 0.35, None, 'soft', 2.0)) # frozen: below 0.35 x the refs' median
+# key bad kind margin scale (refs only; dB margins widen to a quarter of the refs' range)
+POLISH = (('attack_ms', +1, 'log', 1.5, 1.0), # slower than 1.5 x the refs' slowest (+1 ms)
+ ('onset_hf_db', -1, 'db', 4.0, 6.0), # duller snap than the refs' dullest - 4 dB
+ ('bump_db', +1, 'db', 2.0, 6.0), # bumpier tail than the refs' bumpiest + 2 dB (abs 4)
+ ('t40_ms', -1, 'log', 0.6, 1.0), # tail shorter than 0.6 x the refs' shortest
+ ('crest_db', +1, 'db', 2.0, 4.0)) # thinner than the refs' thinnest + 2 dB
+HARD_AT = 0.5
+
+
+class PolishRef:
+ """the refs' artifact and polish numbers, and a check of any sound against them (or absolute limits alone)"""
+
+ def __init__(self, refs=(), designed=False):
+ self.M = [measure(_load(r)) for r in refs]
+ self.n = len(self.M)
+ self.designed = designed
+
+ def _vals(self, key, keep):
+ v = [self.M[j].get(key) for j in keep]
+ return [float(a) for a in v if a is not None]
+
+ def check(self, x, exclude=None, G=None, m=None):
+ """measure one sound (or pass its measure() dict as m) and score it; exclude = a ref index to leave out"""
+ m = measure(x, G) if m is None else m
+ keep = [j for j in range(self.n) if j != exclude]
+ pens, limits, hard = {}, {}, []
+ for key, bad, ab, margin, scale, kind, cap in RULES:
+ v = m.get(key)
+ if v is None:
+ continue
+ rv = self._vals(key, keep)
+ if bad < 0: # frozen: relative to moving refs only
+ if len(rv) < 2 or np.median(rv) < 1.5:
+ continue
+ lim = margin * float(np.median(rv))
+ p = min(cap, max(0.0, (lim - v) / max(lim, 1e-6) * 2))
+ else:
+ lim = max(ab, max(rv) + margin) if rv else ab
+ p = min(cap, max(0.0, (v - lim) / scale))
+ if self.designed and key in ('clicks', 'silence_ms'):
+ kind, p = 'soft', 0.5 * p
+ limits[key] = round(lim, 2)
+ if p > 0:
+ pens[key] = round(p, 3)
+ if kind == 'hard' and p >= HARD_AT:
+ hard.append(key)
+ art = sum(pens.values())
+ ppen = {}
+ for key, bad, kind, margin, scale in POLISH:
+ v = m.get(key)
+ rv = self._vals(key, keep)
+ if v is None:
+ continue
+ if key == 'bump_db':
+ lim = max(4.0, max(rv) + margin) if rv else 4.0
+ p = (v - lim) / scale
+ elif len(rv) < 2:
+ continue
+ elif kind == 'log':
+ lim = margin * max(rv) + 1.0 if bad > 0 else margin * min(rv)
+ p = np.log2(max(v, 1e-3) / lim) if bad > 0 else np.log2(lim / max(v, 1e-3))
+ p /= scale
+ else:
+ mg = max(margin, 0.25 * (max(rv) - min(rv)))
+ lim = max(rv) + mg if bad > 0 else min(rv) - mg
+ p = (v - lim) / scale if bad > 0 else (lim - v) / scale
+ limits[key] = round(float(lim), 2)
+ if p > 0:
+ ppen[key] = round(float(min(2.0, p)), 3)
+ out = {'measures': m, 'limits': limits, 'art': pens, 'polish': ppen, 'hard': hard,
+ 'art_pen': round(float(art), 3), 'polish_pen': round(float(sum(ppen.values())), 3)}
+ out['words'] = self.words(out, keep)
+ return out
+
+ def words(self, c, keep=None):
+ m, keep = c['measures'], keep if keep is not None else range(self.n)
+
+ def rng(key, fmt='{:.0f}'):
+ rv = self._vals(key, keep)
+ return f" (refs {fmt.format(min(rv))}-{fmt.format(max(rv))})" if rv else ''
+ say = {'clicks': lambda: f"{m['clicks']} click{'s' if m['clicks'] != 1 else ''} "
+ f"({', '.join(f'{t:.3f} s' for t in m['click_times'][:3])})",
+ 'edge_db': lambda: f"starts / ends off zero ({m['edge_db']:.0f} dB re peak)",
+ 'fold': lambda: f"aliasing: {m['fold']:.0%} of the moving partials fold back",
+ 'nyq_db': lambda: f"energy above 20 kHz {m['nyq_db']:.0f} dB" + rng('nyq_db'),
+ 'dc_db': lambda: f"DC offset {m['dc_db']:.0f} dB re peak",
+ 'infra_db': lambda: f"infrasound below 15 Hz {m['infra_db']:.0f} dB" + rng('infra_db'),
+ 'res_db': lambda: f"narrow resonance at {m['res_hz']} Hz +{m['res_db']:.0f} dB" + rng('res_db'),
+ 'ring_z': lambda: f"metallic comb ring every {m['ring_ms']:.1f} ms (z {m['ring_z']:.0f})" + rng('ring_z'),
+ 'trunc_db': lambda: f"tail cut at {m['trunc_db']:.0f} dB",
+ 'silence_ms': lambda: f"{m['silence_ms']:.0f} ms of digital silence inside",
+ 'clip_runs': lambda: f"clipped ({m['clip_runs']} flat tops)",
+ 'evolution_db': lambda: (f"frozen spectrum ({m['evolution_db']:.1f} dB of change, refs median "
+ f"{np.median(self._vals('evolution_db', keep)):.1f})"),
+ 'attack_ms': lambda: f"attack {m['attack_ms']:.0f} ms" + rng('attack_ms', '{:.1f}'),
+ 'onset_hf_db': lambda: f"dull onset (>6 kHz share {m['onset_hf_db']:.0f} dB" + rng('onset_hf_db') + ')',
+ 'bump_db': lambda: f"bumpy tail (+{m['bump_db']:.0f} dB re-rise)",
+ 't40_ms': lambda: f"tail dies fast (t40 {m['t40_ms']:.0f} ms" + rng('t40_ms') + ')',
+ 'crest_db': lambda: f"thin body (crest {m['crest_db']:.1f} dB" + rng('crest_db', '{:.1f}') + ')'}
+ a = [say[k]() for k in sorted(c['art'], key=lambda k: -c['art'][k])]
+ p = [say[k]() for k in sorted(c['polish'], key=lambda k: -c['polish'][k])]
+ bits = (['artifacts: ' + ', '.join(a)] if a else []) + (['polish: ' + ', '.join(p)] if p else [])
+ return '; '.join(bits) if bits else 'clean (no artifact or polish flag)'
+
+
+if __name__ == '__main__': # python -m earshot.forge.polish files/folders
+ import sys
+ from .. import audio
+ for p in audio.list_media(sys.argv[1:], exts=audio.AUDIO_EXTS):
+ r = measure(p)
+ print(p.replace(chr(92), '/').rsplit('/', 1)[-1], ' '.join(f'{k}={r[k]}' for k in
+ ('clicks', 'edge_db', 'fold', 'nyq_db', 'dc_db', 'res_db', 'res_hz', 'ring_z', 'trunc_db', 'silence_ms',
+ 'clip_runs', 'evolution_db', 'attack_ms', 'onset_hf_db', 'bump_db', 't40_ms', 'crest_db')))
diff --git a/earshot/forge/recipes/click.py b/earshot/forge/recipes/click.py
index 26fb01e..c3e73fd 100644
--- a/earshot/forge/recipes/click.py
+++ b/earshot/forge/recipes/click.py
@@ -51,6 +51,7 @@ from . import REF_ROOT
TARGET = 'a soft premium UI button click'
ALTS = ['a mechanical keyboard key press', 'a wooden knock', 'a camera shutter click', 'a finger snap']
_K = REF_ROOT + '/ui/'
+REFSET = 'click' # the CLI's default refs: this refsets category, 'forge' split (bench held out)
REFS = [_K + f for f in ('ui_click_01.wav', 'ui_click_02.wav', 'ui_click_03.wav', 'ui_click_04.wav',
'ui_toggle_01.wav', 'ui_toggle_02.wav')]
diff --git a/earshot/forge/recipes/impact.py b/earshot/forge/recipes/impact.py
index 0735793..ae0cedd 100644
--- a/earshot/forge/recipes/impact.py
+++ b/earshot/forge/recipes/impact.py
@@ -61,6 +61,7 @@ SR = dsp.SR
LO = 6000 # sample rate of the low layers (body, sub, rumble), up-sampled x8
TARGET = 'a deep cinematic boom impact'
ALTS = ['a kick drum', 'an explosion', 'a door slam', 'thunder']
+REFSET = 'impact' # the CLI's default refs: this refsets category, 'forge' split (bench held out)
REFS = [f'{REF_ROOT}/impact/impact_0{i}.wav' for i in (1, 2, 3)]
MEMBRANE = [1.0, 1.594, 2.136, 2.296, 2.653, 2.918]
diff --git a/earshot/forge/recipes/keypress.py b/earshot/forge/recipes/keypress.py
index b38af73..3a7925c 100644
--- a/earshot/forge/recipes/keypress.py
+++ b/earshot/forge/recipes/keypress.py
@@ -43,6 +43,7 @@ SR = dsp.SR
TARGET = 'a single deep thocky mechanical keyboard key press'
ALTS = ['a wooden knock', 'a computer mouse click', 'a typewriter key', 'a cork pop']
_K = REF_ROOT + '/keyboard/'
+REFSET = 'typing_key' # the CLI's default refs: this refsets category, 'forge' split (bench held out)
REFS = [_K + f for f in ('kb_thock_01.wav', 'kb_thock_02.wav', 'kb_thock_03.wav', 'kb_thock_04.wav',
'kb_enter_01.wav', 'kb_space_01.wav')]
diff --git a/earshot/forge/recipes/pop.py b/earshot/forge/recipes/pop.py
index 4aabfa9..2710f0a 100644
--- a/earshot/forge/recipes/pop.py
+++ b/earshot/forge/recipes/pop.py
@@ -46,6 +46,7 @@ from . import REF_ROOT
TARGET = 'a small bubble pop'
ALTS = ['a cork pop', 'a water drop', 'a finger snap', 'a mouse click']
+REFSET = 'pop' # the CLI's default refs: this refsets category, 'forge' split (bench held out)
REFS = [REF_ROOT + '/ui/ui_pop_0%d.wav' % i for i in (1, 2, 3, 4)]
SPACE = {
diff --git a/earshot/forge/recipes/whoosh.py b/earshot/forge/recipes/whoosh.py
index 75e1a71..d0a9b0b 100644
--- a/earshot/forge/recipes/whoosh.py
+++ b/earshot/forge/recipes/whoosh.py
@@ -54,6 +54,7 @@ TARGET = 'a fast crisp air whoosh'
ALTS = ['a wind gust', 'rain', 'a sword swing', 'a jet engine']
# round 12: the fast references only (the target is a FAST whoosh; rounds 1-11 also used whoosh_med_01-03, whose
# longer tails the critic then rewarded). REFS_ROUNDS_1_11 is the old set, kept for reproducing those reports.
+REFSET = 'whoosh_fast' # the CLI's default refs: this refsets category, 'forge' split (bench held out)
REFS = [f'{REF_ROOT}/whoosh/whoosh_fast_0{i}.wav' for i in range(1, 5)]
REFS_ROUNDS_1_11 = REFS + [f'{REF_ROOT}/whoosh/whoosh_med_0{i}.wav' for i in range(1, 4)]
diff --git a/earshot/forge/refsets.py b/earshot/forge/refsets.py
new file mode 100644
index 0000000..70946bc
--- /dev/null
+++ b/earshot/forge/refsets.py
@@ -0,0 +1,164 @@
+"""Reference sets for the forge: per category, real / pro recordings split by source file into 'forge' and 'bench'.
+
+ from earshot.forge.refsets import refset, categories, note
+ refs = refset('whoosh_fast') # the forge may optimise against these (Critic refs=, ToneRef)
+ held = refset('whoosh_fast', split='bench') # held out: only eval/forge/bench.py reads these, never the forge
+ categories() # ['click', 'downlifter', 'glitch', 'hover_tick', 'impact', ...]
+ note('sub_drop') # why a set is thin or empty ('' when nothing to say)
+ held_out(paths) # which of these must never be forge refs (bench / bench-source kit)
+
+The recordings are licensed third-party sounds and are not redistributed, so the sets live in a local registry, a JSON
+file: EARSHOT_REFSETS, else <EARSHOT_HOME>/refsets.json. On the dev machine eval/forge/build_refsets.py writes it from
+the licensed library (trims located with earshot find / rank, trim + fade + peak gain only). Public users bring their
+own refs: lay them out as <root>/<category>/{forge,bench}/*.wav and run `write(scan(root))`, or write the JSON by hand:
+ {"version": 1, "categories": {"whoosh_fast": {"forge": ["a.wav", ...], "bench": ["b.wav", ...], "note": ""}}}
+Relative paths are read relative to the registry file.
+
+The split rule that makes the benchmark mean something: no source recording (nor a sister take of it) is in both
+splits of any category, so a forge tuned against 'forge' refs is measured against recordings it never heard.
+CATEGORIES lists the motion-design SFX categories the benchmark knows, with what each one means.
+"""
+import json
+import os
+
+SPLITS = ('forge', 'bench')
+CATEGORIES = {
+ 'whoosh_fast': 'a short crisp pass-by whoosh, a stick or rod swung through air (0.2-0.8 s)',
+ 'whoosh_big': 'a big, deep, slower whoosh / swoop (0.8-3 s)',
+ 'swish': 'a light thin quick swish, a thrown object (0.15-0.5 s)',
+ 'swipe': 'a soft cloth / paper slide or swipe (0.25-1 s)',
+ 'riser': 'a swell building up to a peak (1.5-6 s)',
+ 'downlifter': 'a falling sweep that sinks away after a hit (1-4 s)',
+ 'impact': 'a deep cinematic boom hit with a long tail (1-8 s)',
+ 'sub_drop': 'a sub-bass pitch drop felt under a hit (1-4 s)',
+ 'click': 'a small plastic UI button click (30-350 ms)',
+ 'tap': 'a soft finger tap on glass / a hard surface (30-300 ms)',
+ 'toggle': 'a switch flipped: a crisp click, often two-part (50-450 ms)',
+ 'hover_tick': 'a tiny soft tick for hovers and scroll detents (15-150 ms)',
+ 'pop': 'a short round pop: bubble, mouth, cork (50-500 ms)',
+ 'notify': 'a short bright notification ding / chime (0.4-2 s)',
+ 'shimmer': 'a sparkling chime sweep / bell-tree shimmer (1-6 s)',
+ 'typing_key': 'a single mechanical keyboard key press (60-450 ms)',
+ 'glitch': 'a digital glitch / stutter burst (designed; 0.1-1 s)',
+ 'logo_sting': 'a short logo sting: a struck hit with a ringing tail (1-6 s)',
+}
+
+
+# kit files (the v1 recipes' REFS, refs/<group>/<name>.wav) cut from sources that are 'bench' in some category: a forge
+# run must never use them (they would leak the held-out side into the search). held_out() flags them.
+KIT_BENCH = ('ui_notify_02', 'whoosh_med_01', 'whoosh_med_02', 'whoosh_big_02', 'swipe_01', 'riser_02', 'impact_03',
+ 'shimmer_01')
+
+
+class RefsetsMissing(FileNotFoundError):
+ """no reference-set registry on this machine"""
+
+
+def registry_path():
+ """where the registry is read from: EARSHOT_REFSETS, else <EARSHOT_HOME>/refsets.json"""
+ if os.environ.get('EARSHOT_REFSETS'):
+ return os.path.abspath(os.environ['EARSHOT_REFSETS'])
+ from .. import config
+ return os.path.join(config.HOME, 'refsets.json')
+
+
+def load(path=None):
+ """the registry dict ({'version', 'categories': {cat: {'forge', 'bench', 'note'}}}) with absolute paths"""
+ path = path or registry_path()
+ if not os.path.exists(path):
+ raise RefsetsMissing(
+ f'no reference-set registry at {path}. The forge refs are licensed recordings and are not shipped: bring '
+ 'your own (lay them out as <root>/<category>/{forge,bench}/*.wav and run '
+ 'earshot.forge.refsets.write(scan(root))), point EARSHOT_REFSETS at a registry JSON, or on the dev '
+ 'machine run eval/forge/build_refsets.py.')
+ with open(path, encoding='utf-8') as fh:
+ reg = json.load(fh)
+ base = os.path.dirname(os.path.abspath(path))
+ for c in reg.get('categories', {}).values():
+ for s in SPLITS:
+ c[s] = [p if os.path.isabs(p) else os.path.normpath(os.path.join(base, p)).replace(chr(92), '/')
+ for p in c.get(s, [])]
+ return reg
+
+
+def categories(path=None):
+ """category names present in the registry (sorted)"""
+ return sorted(load(path).get('categories', {}))
+
+
+def _cat(category, path):
+ cats = load(path).get('categories', {})
+ if category not in cats:
+ raise KeyError(f"no reference set '{category}' in {path or registry_path()}; known: {', '.join(sorted(cats))}")
+ return cats[category]
+
+
+def refset(category, split='forge', path=None):
+ """paths of the `split` ('forge' | 'bench') refs of `category` (may be empty: see note()); missing files raise"""
+ if split not in SPLITS:
+ raise ValueError(f"split must be one of {SPLITS}, not {split!r}")
+ paths = list(_cat(category, path).get(split, []))
+ gone = [p for p in paths if not os.path.exists(p)]
+ if gone:
+ raise RefsetsMissing(f"{len(gone)} {category}/{split} ref(s) missing, e.g. {gone[0]}: rebuild the refs")
+ return paths
+
+
+def note(category, path=None):
+ """the registry's note on a category ('' when there is none)"""
+ return _cat(category, path).get('note', '')
+
+
+def held_out(paths, kit=True, path=None):
+ """the paths among `paths` that must not be forge refs: a file in any category's 'bench' split of the registry
+ (when there is one) or, with kit=True, a kit file cut from a bench source (KIT_BENCH, matched by file stem)"""
+ norm = lambda q: os.path.normcase(os.path.abspath(q))
+ try:
+ bench = {norm(q) for c in load(path).get('categories', {}).values() for q in c.get('bench', [])}
+ except RefsetsMissing:
+ bench = set()
+ return [q for q in paths if norm(q) in bench
+ or (kit and os.path.splitext(os.path.basename(q))[0] in KIT_BENCH)]
+
+
+def scan(root, exts=('.wav', '.flac', '.ogg', '.mp3', '.aif', '.aiff')):
+ """a registry dict from a folder layout <root>/<category>/{forge,bench}/<files> (absolute paths)"""
+ cats = {}
+ for c in sorted(os.listdir(root)):
+ d = os.path.join(root, c)
+ if not os.path.isdir(d) or c.startswith(('_', '.')):
+ continue
+ entry = {s: sorted(os.path.join(d, s, f).replace(chr(92), '/') for f in os.listdir(os.path.join(d, s))
+ if f.lower().endswith(exts)) if os.path.isdir(os.path.join(d, s)) else [] for s in SPLITS}
+ if entry['forge'] or entry['bench']:
+ cats[c] = {**entry, 'note': ''}
+ return {'version': 1, 'categories': cats}
+
+
+def check_split(reg, sources=None):
+ """problems with a registry as a list of strings (empty = fine): a file in both splits of a category, or (with
+ sources = {path: source_id}) a source id in forge anywhere and in bench anywhere"""
+ bad = []
+ seen = {s: {} for s in SPLITS}
+ for c, e in reg.get('categories', {}).items():
+ both = set(e.get('forge', [])) & set(e.get('bench', []))
+ bad += [f'{c}: {os.path.basename(p)} in both splits' for p in sorted(both)]
+ for s in SPLITS:
+ for p in e.get(s, []):
+ if sources and p in sources:
+ seen[s].setdefault(sources[p], set()).add(c)
+ for src in sorted(set(seen['forge']) & set(seen['bench'])):
+ bad.append(f"source {src} is forge in {sorted(seen['forge'][src])} and bench in {sorted(seen['bench'][src])}")
+ return bad
+
+
+def write(reg, path=None):
+ """write a registry dict (from scan() or by hand) to `path` (default registry_path()); returns the path"""
+ path = path or registry_path()
+ bad = check_split(reg)
+ if bad:
+ raise ValueError('refuse to write a registry that leaks across splits: ' + '; '.join(bad[:3]))
+ os.makedirs(os.path.dirname(os.path.abspath(path)), exist_ok=True)
+ with open(path, 'w', encoding='utf-8') as fh:
+ json.dump(reg, fh, indent=1)
+ return path
diff --git a/earshot/forge/search.py b/earshot/forge/search.py
index e98bbdf..acc5def 100644
--- a/earshot/forge/search.py
+++ b/earshot/forge/search.py
@@ -1,32 +1,58 @@
-"""The listen-and-improve loop: an evolutionary search over a recipe's parameters, scored by the Critic.
+"""The listen-and-improve loop: a search over a recipe's parameters, scored by the Critic.
from earshot.forge.search import optimize
- res = optimize(recipe, critic, budget=160, pop=16, seed=0)
- res['best'][0]['params'], res['best'][0]['score']['summary'], res['history']
+ res = optimize(recipe, critic, budget=160, pop=16, seed=0) # v1 evolution strategy
+ res = optimize(recipe, critic, budget=160, pop=16, strategy='cma', prescreen=48) # v2: CMA-ES + model-free pre-screen
+ res['best'][0]['params'], res['best'][0]['score']['summary'], res['history'], res['gens'], res['timing']
A recipe is a module (or object) with
SPACE {name: (lo, hi)} | {name: (lo, hi, 'log')} | {name: (lo, hi, 'int')} | {name: [choice, ...]}
DEFAULTS {name: value} (a sensible starting sound; optional)
render(params, r) -> np.ndarray (48 kHz mono or stereo float; r = numpy Generator for natural variation)
-Each generation renders `pop` candidates (elites mutated with a shrinking step, plus a few fresh random ones), peak-
-normalises them to -1 dBFS (so level never decides), scores them in one batch, and keeps the best. Deterministic for a
-given seed. Every candidate is kept in the history, so an agent can read what helped.
+Candidates are peak-normalised to -1 dBFS (so level never decides) and scored in batches. Deterministic for a given
+seed (renders run on `workers` threads, default 2, each on its own seed: the same samples as a serial run). Every
+scored candidate is kept in the history, so an agent can read what helped.
+
+Strategies (`budget` always counts critic-scored candidates):
+ 'es' (v1, the default) elites mutated with a shrinking step (sigma0 -> sigma1) plus a few fresh random ones.
+ 'cma' separable CMA-ES in the unit cube (diagonal covariance, mirrored at the bounds) with restarts: when the step
+ collapses or the best has not moved for `patience` generations it restarts from the strongest unexplored
+ elite, then from random points, with the step reset (initial step 0.15, `cma=` options).
+Measured (keypress, 2 render threads): at 96-240 scored candidates CMA did NOT beat the v1 ES - on a model-free proxy
+objective (the pre-screen score, 3-5 seeds) ES -1.23 vs CMA -1.26..-1.36 at 96, -1.136 vs -1.154 at 240; with the real
+critic (2 seeds, 96) best 6.55 vs 6.03, top-5 mean 5.77 vs 5.30. Elitist and small-mu CMA variants were worse. The
+pre-screen is what helps: ES + prescreen 48 raised the passing candidates from 10.5 to 36.5 of 96 and the top-5 mean
+from 5.77 to 6.17 (CMA + prescreen: 25.5 and 5.92). So 'es' + prescreen is the default the CLI uses.
+`prescreen=K` (K > pop): render K candidates per generation, rank them with model-free measurements against the
+critic's reference recordings (Prescreen: 1/3-octave tonal balance in units of the refs' own spread, envelope stats,
+artifact checks) and embed only the best `pop` with CLAP. The model checks are the expensive part (CLAP is ~70 % of a
+generation), the measurements cost ~10 ms a sound, and they are the signal that still separates made from real once
+the model checks saturate. `reps=N` renders each scored candidate on N seeds and ranks it by the mean (robust to lucky
+renders). Starting points: the recipe's DEFAULTS, `init` (e.g. a seed bank, `load_seeds()`) are scored in generation 0.
`space=` restricts the search to a family (a narrower {name: spec} dict; parameters it leaves out keep the recipe's
-DEFAULTS). `robust()` re-renders a result on new noise seeds: the best single render of a search is a winner's curse
-(the forge makers measured robust means 0.1-2.2 below the reported best), so judge finals by their robust mean.
-`reps=N` fights that inside the search: every candidate is rendered on N seeds and ranked by its mean total (it passes
-when at least half of its renders pass); the row kept is the first render's, with 'total_reps' listing all N.
+DEFAULTS). `robust()` re-renders a result on new noise seeds (the best single render of a search is a winner's curse:
+judge finals by their robust mean). `diverse()` picks strong results that differ in parameters AND in sound (tonal
+balance + envelope). The seed bank (earshot/forge/seeds/<recipe>.json, small, committed) keeps good parameter sets
+found so far: `load_seeds()` / `save_seeds()`.
"""
+import json
import math
+import os
import sys
+import time
import traceback
+from concurrent.futures import ThreadPoolExecutor
import numpy as np
from . import dsp
+SR = 48000
+SEEDS_DIR = os.path.join(os.path.dirname(os.path.abspath(__file__)), 'seeds')
+
+# ---------------------------------------------------------------------------------------------- parameters
def _decode(space, u):
p = {}
for (k, spec), v in zip(space.items(), u):
@@ -63,6 +89,22 @@ def _encode(space, params):
return np.clip(np.array(u, float), 0, 1)
+def _inside(spec, v):
+ if isinstance(spec, list):
+ return v in spec
+ try:
+ return spec[0] <= v <= spec[1] and (len(spec) < 3 or spec[2] != 'int' or float(v).is_integer())
+ except TypeError:
+ return False
+
+
+def _mirror(x):
+ """fold any real vector into [0, 1] by mirroring at the bounds (keeps the sampling distribution's shape)"""
+ r = np.mod(x, 2.0)
+ return np.where(r > 1.0, 2.0 - r, r)
+
+
+# ---------------------------------------------------------------------------------------------- rendering
_failed = set()
@@ -84,88 +126,554 @@ def render(recipe, params, seed):
return dsp.normalize(x, -1.0).astype(np.float32)
+def _workers(workers):
+ if workers is None:
+ workers = int(os.environ.get('EARSHOT_THREADS') or 2)
+ return max(1, min(int(workers), 8))
+
+
+def render_many(recipe, jobs, workers=None):
+ """render [(params, seed), ...] on `workers` threads (default EARSHOT_THREADS or 2); same order and samples as
+ a serial loop (each render owns its seed). Returns a list of arrays / None."""
+ jobs = list(jobs)
+ w = _workers(workers)
+ if w == 1 or len(jobs) < 2:
+ return [render(recipe, p, s) for p, s in jobs]
+ with ThreadPoolExecutor(w) as ex:
+ return list(ex.map(lambda j: render(recipe, j[0], j[1]), jobs))
+
+
+# ---------------------------------------------------------------------------------------------- model-free ears
+ENV_KEYS = ('attack_ms', 't10_ms', 't20_ms', 't40_ms', 'centroid_ms', 'crest_db', 'end_db')
+_DB_KEYS = 2 # the last two are levels in dB (6 dB units), the rest log times
+
+
+def _mono(x):
+ x = np.asarray(x, np.float64)
+ return x.mean(axis=1) if x.ndim == 2 else x
+
+
+def envelope_stats(x, hop_ms=2.0):
+ """envelope measurements of one sound (mono sum): attack (onset at -30 dB to peak), t10 / t20 / t40 (peak until the
+ 2 ms RMS envelope first falls 10 / 20 / 40 dB under it, or the file end), temporal centroid from the onset, and
+ the crest factor (peak / RMS of the part within 40 dB of the peak), end_db (mean level of the last quarter of the
+ file re the peak, floored at -90: a noise-floor plateau where real edits fade to silence), all in ms or dB"""
+ m = _mono(x)
+ n = max(1, int(SR * hop_ms / 1000))
+ k = len(m) // n
+ if k < 2:
+ return {key: (-90.0 if key == 'end_db' else 0.0) for key in ENV_KEYS}
+ pw = (m[:k * n].reshape(k, n) ** 2).mean(1)
+ e = 10 * np.log10(pw + 1e-18)
+ ip = int(np.argmax(e))
+ pk = e[ip]
+ on = int(np.argmax(e > pk - 30))
+ out = {'attack_ms': (ip - on) * hop_ms + hop_ms / 2}
+ for db in (10, 20, 40):
+ idx = np.nonzero(e[ip:] < pk - db)[0]
+ out[f't{db}_ms'] = float((idx[0] if len(idx) else k - ip) * hop_ms)
+ w = pw[on:]
+ out['centroid_ms'] = float((np.arange(len(w)) * w).sum() / (w.sum() + 1e-18) * hop_ms + hop_ms / 2)
+ act = m[on * n:k * n][np.repeat(e[on:] > pk - 40, n)]
+ out['crest_db'] = float(20 * np.log10(np.abs(m).max() / (np.sqrt((act ** 2).mean()) + 1e-12) + 1e-12)) \
+ if len(act) else 0.0
+ out['end_db'] = float(max(-90.0, 10 * np.log10(pw[int(0.75 * k):].mean() + 1e-18) - pk))
+ return out
+
+
+def artifacts(x):
+ """cheap artifact checks on one sound: {'flags': {name: value}, 'penalty': float}. flat_top (clipped runs at the
+ peak), dc (mean offset vs RMS), start_jump (the first sample already loud: an unfaded click), cut_end (the last
+ 5 ms within 45 dB of the peak: a tail cut off), gap (digital silence inside the sound), ultrasonic (energy share
+ above 20 kHz). The fallback when earshot.forge.polish (finer detectors, ref-relative) is not available."""
+ m = _mono(x)
+ flags, pen = {}, 0.0
+ pk = np.abs(m).max() + 1e-12
+ hot = np.abs(m) > 0.98 * pk
+ if hot.sum() >= 4:
+ d = np.diff(np.concatenate([[0], hot.astype(np.int8), [0]]))
+ runs = np.nonzero(d == -1)[0] - np.nonzero(d == 1)[0]
+ if runs.max() >= 4:
+ flags['flat_top'] = int(runs.max())
+ pen += 1.0
+ rms = np.sqrt((m ** 2).mean()) + 1e-12
+ if abs(m.mean()) / rms > 0.1:
+ flags['dc'] = round(float(m.mean() / rms), 3)
+ pen += 0.5
+ if abs(m[0]) / pk > 0.05:
+ flags['start_jump'] = round(float(abs(m[0]) / pk), 3)
+ pen += 0.5
+ tail = m[-int(0.005 * SR):]
+ end_db = 20 * np.log10(np.sqrt((tail ** 2).mean()) / pk + 1e-12)
+ if end_db > -45:
+ flags['cut_end'] = round(float(end_db), 1)
+ pen += 1.0
+ z = (m == 0).astype(np.int8)
+ if z.any():
+ d = np.diff(np.concatenate([[0], z, [0]]))
+ st, en = np.nonzero(d == 1)[0], np.nonzero(d == -1)[0]
+ inner = [(a, b) for a, b in zip(st, en) if a > 0 and b < len(m) and b - a > 0.005 * SR]
+ if inner:
+ flags['gap'] = len(inner)
+ pen += 1.0
+ F = np.abs(np.fft.rfft(m)) ** 2
+ f = np.fft.rfftfreq(len(m), 1 / SR)
+ us = 10 * np.log10(F[f > 20000].sum() / (F.sum() + 1e-18) + 1e-12)
+ if us > -25:
+ flags['ultrasonic'] = round(float(us), 1)
+ pen += 0.5
+ return {'flags': flags, 'penalty': round(pen, 3)}
+
+
+def describe(x):
+ """the model-free descriptor of one sound: 1/3-octave profile (earshot.forge.tone), envelope stats, artifacts"""
+ from .tone import profile
+ return {'profile': profile(x), 'env': envelope_stats(x), 'art': artifacts(x)}
+
+
+def _envvec(env):
+ return np.array([math.log(max(env[k], 0.5)) for k in ENV_KEYS[:-_DB_KEYS]]
+ + [env.get(k, -90.0 if k == 'end_db' else 0.0) / 6.0 for k in ENV_KEYS[-_DB_KEYS:]])
+
+
+def sound_dist(a, b):
+ """distance between two descriptors in dB-like units: RMS 1/3-octave difference plus 4 dB per e-fold of envelope
+ timing difference (RMS over attack / t10 / t20 / t40 / centroid, crest and end level in 6 dB units)"""
+ pa, pb = a['profile'], b['profile']
+ return float(np.sqrt(np.mean((pa - pb) ** 2)) + 4.0 * np.sqrt(np.mean((_envvec(a['env']) - _envvec(b['env'])) ** 2)))
+
+
+class Prescreen:
+ """ranks sounds against reference recordings with measurements only (no model), cheap enough to run on many
+ renders before the critic embeds a few. score(xs) -> rows {'pre', 'tone_db', 'tone_ratio', 'env_pen', 'env_z',
+ 'art', 'desc'[, 'polish']}; pre (higher is better) = -(tone_ratio + env_w * env_pen + artifacts), env_pen = mean
+ over env_keys of the robust z beyond 1.5 (both directions: a tail that dies too fast counts too). artifacts =
+ earshot.forge.polish's art_pen + 0.75 polish_pen + 1 per hard artifact, as the critic weighs them (`polish_ref=`
+ the critic's PolishRef, else one built from the refs; ~45 ms a sound), or this module's artifacts() without polish.
+ Measured against the critic (tone=1.5) on random candidates around DEFAULTS: without polish, Spearman 0.60-0.63
+ (keypress, 96), picking 16 of 48 raised the mean critic total in all four halves tested; with polish vs critic
+ v2 (72 candidates, 12 of 36 picked): whoosh Spearman 0.53 -> 0.69, selected mean 6.66 -> 7.71 and 4.35 -> 6.02
+ (critic's best kept in both halves); keypress 0.60 -> 0.71, 1.43 -> 2.40 and -1.99 -> 0.56 (best kept in one
+ half). Without refs only the artifact checks rank."""
+
+ def __init__(self, refs=None, tone_target=None, env_w=0.5, env_keys=('t40_ms', 'crest_db', 'end_db'),
+ polish_ref=None, polish=True):
+ from .. import audio
+ xs = [audio.load(r) if isinstance(r, str) else np.asarray(r) for r in (refs or [])]
+ self.pol = polish_ref
+ if self.pol is None and polish:
+ try:
+ from .polish import PolishRef
+ self.pol = PolishRef(xs)
+ except Exception: # polish missing or broken: the built-in checks rank alone
+ self.pol = None
+ self.n = len(xs)
+ self.tone_target, self.env_w = tone_target, float(env_w)
+ self.env_idx = [ENV_KEYS.index(k) for k in env_keys]
+ self.descs = [describe(x) for x in xs]
+ if self.n:
+ P = np.stack([d['profile'] for d in self.descs])
+ self.med = np.median(P, 0)
+ if self.n >= 3:
+ loo = [float(np.sqrt(np.mean((P[i] - np.median(np.delete(P, i, 0), 0)) ** 2))) for i in range(self.n)]
+ self.spread = max(float(np.median(loo)), 0.5)
+ else:
+ self.spread = 4.0
+ if tone_target is not None:
+ self.spread = float(tone_target)
+ V = np.stack([_envvec(d['env']) for d in self.descs])
+ self.env_med = np.median(V, 0)
+ q1, q3 = np.percentile(V, [25, 75], axis=0)
+ floor = np.array([0.2] * (len(ENV_KEYS) - _DB_KEYS) + [2.0 / 6.0] * _DB_KEYS)
+ self.env_sd = np.maximum((q3 - q1) / 1.35, floor)
+
+ def row(self, x, polish=True):
+ d = describe(x)
+ r = {'art': d['art'], 'desc': d}
+ art = d['art']['penalty']
+ if polish and self.pol is not None:
+ try:
+ pm = self.pol.check(x)
+ r['polish'] = {'art_pen': round(float(pm['art_pen']), 3), 'polish_pen': round(float(pm['polish_pen']), 3),
+ 'hard': list(pm.get('hard') or [])}
+ art = r['polish']['art_pen'] + 0.75 * r['polish']['polish_pen'] + 1.0 * len(r['polish']['hard'])
+ except Exception:
+ pass
+ if self.n:
+ r['tone_db'] = round(float(np.sqrt(np.mean((d['profile'] - self.med) ** 2))), 2)
+ r['tone_ratio'] = round(r['tone_db'] / self.spread, 3)
+ z = (_envvec(d['env']) - self.env_med) / self.env_sd
+ r['env_z'] = {k: round(float(v), 2) for k, v in zip(ENV_KEYS, z)}
+ r['env_pen'] = round(float(np.mean(np.maximum(0.0, np.abs(z[self.env_idx]) - 1.5))), 3) if self.env_idx else 0.0
+ r['pre'] = round(-(r['tone_ratio'] + self.env_w * r['env_pen'] + art), 3)
+ else:
+ r['pre'] = round(-art, 3)
+ return r
+
+ def score(self, xs, polish=True):
+ return [self.row(x, polish) if x is not None else {'pre': -np.inf, 'art': {'flags': {'silent': 1}, 'penalty': 9.0}}
+ for x in xs]
+
+ def nearest(self, desc):
+ """index of the nearest reference to a descriptor (sound_dist), or None without refs"""
+ if not self.n:
+ return None
+ return int(np.argmin([sound_dist(desc, r) for r in self.descs]))
+
+
+def prescreen_for(critic):
+ """a Prescreen built from a critic's references (reuses nothing model-based; None-safe for critics without refs)"""
+ return Prescreen(getattr(critic, 'refs', None) or [], tone_target=getattr(critic, 'tone_target', None),
+ polish_ref=getattr(critic, 'polish_ref', None))
+
+
+# ---------------------------------------------------------------------------------------------- the search
+def _evaluate(recipe, critic, params, seeds, reps, workers, first=None):
+ """render (unless `first` holds the first renders) and score candidates; returns ({j: row}, timing dict)"""
+ reps = max(1, reps)
+ t0 = time.time()
+ jobs = [(j, k, s + k * 7919) for j, s in enumerate(seeds) for k in range(reps)]
+ todo = [(params[j], s) for j, k, s in jobs if first is None or k > 0]
+ it = iter(render_many(recipe, todo, workers))
+ xs = [first[j] if (first is not None and k == 0) else next(it) for j, k, _ in jobs]
+ t1 = time.time()
+ live = [i for i, x in enumerate(xs) if x is not None]
+ scores = critic.score([xs[i] for i in live]) if live else []
+ t2 = time.time()
+ got = {}
+ for i, sc in zip(live, scores):
+ got.setdefault(jobs[i][0], []).append((sc, i))
+ by = {}
+ for j, rows in got.items():
+ if len(rows) < reps: # a failed render among the reps: the candidate is unreliable
+ continue
+ if reps > 1:
+ tot = [float(r['total']) for r, _ in rows]
+ row = dict(rows[0][0])
+ row['total_reps'] = [round(t, 3) for t in tot]
+ row['total'] = round(float(np.mean(tot)), 3) if all(np.isfinite(tot)) else -np.inf
+ row['passes'] = sum(1 for r, _ in rows if r.get('passes')) * 2 >= len(rows)
+ by[j] = row
+ else:
+ by[j] = rows[0][0]
+ xs0 = [xs[j * reps] for j in range(len(seeds))]
+ return by, xs0, {'render_s': t1 - t0, 'score_s': t2 - t1, 'rendered': len(todo), 'scored': len(live)}
+
+
+class _CMA:
+ """separable CMA-ES state in the unit cube (Ros & Hansen 2008), mirrored at the bounds"""
+
+ def __init__(self, d, lam, m, sigma, mu_frac=0.5, elite=0.0):
+ self.d, self.lam, self.elite = d, lam, float(elite)
+ mu = max(1, int(round(lam * mu_frac)))
+ w = np.log(mu + 0.5) - np.log(np.arange(1, mu + 1))
+ self.w = w / w.sum()
+ self.mueff = 1.0 / (self.w ** 2).sum()
+ n, me = d, self.mueff
+ self.cs = (me + 2) / (n + me + 5)
+ self.ds = 1 + 2 * max(0.0, math.sqrt((me - 1) / (n + 1)) - 1) + self.cs
+ self.cc = (4 + me / n) / (n + 4 + 2 * me / n)
+ sep = (n + 2) / 3.0
+ self.c1 = min(1.0, sep * 2 / ((n + 1.3) ** 2 + me))
+ self.cmu = min(1 - self.c1, sep * 2 * (me - 2 + 1 / me) / ((n + 2) ** 2 + me))
+ self.chi = math.sqrt(n) * (1 - 1 / (4 * n) + 1 / (21 * n * n))
+ self.reset(m, sigma)
+
+ def reset(self, m, sigma):
+ self.m, self.sigma = np.array(m, float), float(sigma)
+ self.C = np.ones(self.d)
+ self.ps, self.pc = np.zeros(self.d), np.zeros(self.d)
+ self.g = 0
+
+ def ask(self, R, n, elites=()):
+ """n points: around the mean, a fraction `elite` around the best points so far (elitist, for short noisy runs)"""
+ z = R.standard_normal((n, self.d))
+ ne = int(round(self.elite * n)) if len(elites) else 0
+ cen = [elites[i % len(elites)] for i in range(ne)] + [self.m] * (n - ne)
+ return [_mirror(c + self.sigma * np.sqrt(self.C) * zi) for c, zi in zip(cen, z)]
+
+ def tell(self, U, totals):
+ """U: evaluated points (unit cube), totals: their scores (higher is better)"""
+ order = np.argsort(-np.asarray(totals, float), kind='stable')[:len(self.w)]
+ Y = np.array([(U[i] - self.m) / self.sigma for i in order])
+ lim = 2.0 * self.chi # clip injected / mirrored outliers (Hansen 2011)
+ nrm = np.sqrt(((Y / np.sqrt(self.C)) ** 2).sum(1))
+ Y *= np.minimum(1.0, lim / np.maximum(nrm, 1e-12))[:, None]
+ w = self.w[:len(Y)] / self.w[:len(Y)].sum()
+ yw = (w[:, None] * Y).sum(0)
+ self.m = np.clip(self.m + self.sigma * yw, 0.0, 1.0)
+ self.g += 1
+ self.ps = (1 - self.cs) * self.ps + math.sqrt(self.cs * (2 - self.cs) * self.mueff) * yw / np.sqrt(self.C)
+ hs = np.linalg.norm(self.ps) / math.sqrt(1 - (1 - self.cs) ** (2 * self.g)) < (1.4 + 2 / (self.d + 1)) * self.chi
+ self.pc = (1 - self.cc) * self.pc + hs * math.sqrt(self.cc * (2 - self.cc) * self.mueff) * yw
+ self.C = ((1 - self.c1 - self.cmu) * self.C + self.c1 * (self.pc ** 2 + (1 - hs) * self.cc * (2 - self.cc) * self.C)
+ + self.cmu * (w[:, None] * Y ** 2).sum(0))
+ self.C = np.clip(self.C, 1e-4, 16.0)
+ self.sigma = float(np.clip(self.sigma * math.exp((self.cs / self.ds) * (np.linalg.norm(self.ps) / self.chi - 1)),
+ 1e-3, 0.6))
+
+
def optimize(recipe, critic, budget=160, pop=16, elite=4, seed=0, init=None, sigma0=0.22, sigma1=0.04,
- fresh=0.15, log=None, space=None, reps=1):
- """returns {'best': [top results], 'history': [every result], 'generations': n}; a result = {params, seed, score}.
+ fresh=0.15, log=None, space=None, reps=1, strategy='es', prescreen=0, workers=None, patience=4,
+ sigma_stop=0.03, pre=None, cma=None):
+ """returns {'best': [top results], 'history': [every scored result], 'generations': n, 'gens': [per-generation
+ stats], 'timing': {...}, 'strategy', 'prescreen'}; a result = {gen, u, params, seed, score[, pre]}.
space: optional narrower {name: spec} (a family); parameters it leaves out keep the recipe's DEFAULTS.
- reps: renders per candidate (ranked by the mean total; budget counts candidates, so reps=2 costs twice the time)"""
+ reps: renders per candidate (ranked by the mean total; budget counts candidates, so reps=2 costs twice the time).
+ strategy: 'es' (v1) | 'cma' (separable CMA-ES with restarts). prescreen=K > pop: render K per generation, rank
+ them model-free against the critic's refs (`pre=` a prebuilt Prescreen) and score only the best pop.
+ workers: render threads (default EARSHOT_THREADS or 2). patience / sigma_stop: CMA restart triggers; cma= {'sigma0':
+ 0.15 initial / restart step, 'mu': 0.5 parent fraction, 'elite': 0 share of samples around the best so far}."""
base = {}
if space is None:
space = recipe.SPACE
else:
base = dict(getattr(recipe, 'DEFAULTS', {}) or {})
+ full = space is recipe.SPACE
R = np.random.default_rng(seed)
d = len(space)
- starts = [dict(getattr(recipe, 'DEFAULTS', {}) or {})] + [dict(p) for p in (init or [])]
+ defaults = dict(getattr(recipe, 'DEFAULTS', {}) or {})
+ starts = [defaults] + [{**defaults, **dict(p)} for p in (init or [])]
gens = max(1, budget // pop)
- history, elites = [], []
+ K = int(prescreen or 0)
+ use_pre = K > pop
+ if use_pre and pre is None:
+ pre = prescreen_for(critic)
+ history, elites, gstats = [], [], []
+ timing = {'render_s': 0.0, 'pre_s': 0.0, 'score_s': 0.0, 'rendered': 0, 'scored': 0}
+ co = {'sigma0': 0.15, 'mu': 0.5, 'elite': 0.0, **(cma or {})}
+ cma = _CMA(d, pop, _encode(space, starts[0]), co['sigma0'], co['mu'], co['elite']) if strategy == 'cma' else None
+ restarts, best_seen, stale, tried = 0, -np.inf, 0, []
+ t_start = time.time()
for g in range(gens):
- sig = sigma0 * (sigma1 / sigma0) ** (g / max(1, gens - 1))
- U, seeds = [], []
- for j in range(pop):
- if g == 0:
- u = _encode(space, starts[j]) if j < len(starts) else R.uniform(0, 1, d)
- elif R.uniform() < fresh or not elites:
- u = R.uniform(0, 1, d)
- else:
- parent = elites[min(len(elites) - 1, int(R.integers(0, max(1, min(elite, len(elites))))))]
- u = np.clip(parent['u'] + R.normal(0, sig, d), 0, 1)
- U.append(u)
- seeds.append(int(R.integers(0, 2 ** 31)))
+ n_new = K if use_pre else pop
+ U, seeds, fixed = [], [], 0
+ if strategy == 'cma':
+ if g == 0: # the starting points are always scored
+ U = [_encode(space, p) for p in starts[:pop]]
+ fixed = len(U)
+ U += cma.ask(R, n_new - len(U), [h['u'] for h in elites])
+ seeds = [int(R.integers(0, 2 ** 31)) for _ in U]
+ else: # v1 draw order: each seed right after its candidate
+ sig = sigma0 * (sigma1 / sigma0) ** (g / max(1, gens - 1))
+ for j in range(n_new):
+ if g == 0:
+ u = _encode(space, starts[j]) if j < len(starts) and j < pop else R.uniform(0, 1, d)
+ fixed += int(j < len(starts) and j < pop)
+ elif R.uniform() < fresh or not elites:
+ u = R.uniform(0, 1, d)
+ else:
+ parent = elites[min(len(elites) - 1, int(R.integers(0, max(1, min(elite, len(elites))))))]
+ u = np.clip(parent['u'] + R.normal(0, sig, d), 0, 1)
+ U.append(u)
+ seeds.append(int(R.integers(0, 2 ** 31)))
params = [{**base, **_decode(space, u)} for u in U]
- jobs = [(j, s + k * 7919) for j, s in enumerate(seeds) for k in range(max(1, reps))]
- xs = [render(recipe, params[j], s) for j, s in jobs]
- live = [i for i, x in enumerate(xs) if x is not None]
- scores = critic.score([xs[i] for i in live]) if live else []
- got = {}
- for i, sc in zip(live, scores):
- got.setdefault(jobs[i][0], []).append(sc)
- by = {}
- for j, rows in got.items():
- if len(rows) < max(1, reps): # a failed render among the reps: the candidate is unreliable
- continue
- if reps > 1:
- tot = [float(r['total']) for r in rows]
- row = dict(rows[0])
- row['total_reps'] = [round(t, 3) for t in tot]
- row['total'] = round(float(np.mean(tot)), 3) if all(np.isfinite(tot)) else -np.inf
- row['passes'] = sum(1 for r in rows if r.get('passes')) * 2 >= len(rows)
- by[j] = row
- else:
- by[j] = rows[0]
- for j in range(pop):
+ if strategy == 'cma': # starting points keep their exact values (inside the space)
+ for j in range(fixed):
+ params[j].update({k: v for k, v in starts[j].items() if k in space and _inside(space[k], v)})
+ prows, first = None, None
+ if use_pre:
+ t0 = time.time()
+ xs = render_many(recipe, list(zip(params, seeds)), workers)
+ t1 = time.time()
+ prows = pre.score(xs)
+ t2 = time.time()
+ timing['render_s'] += t1 - t0
+ timing['pre_s'] += t2 - t1
+ timing['rendered'] += len(xs)
+ order = list(range(fixed)) + sorted(range(fixed, len(U)), key=lambda i: (-prows[i]['pre'], i))
+ sel = order[:pop]
+ U, seeds, params = [U[i] for i in sel], [seeds[i] for i in sel], [params[i] for i in sel]
+ prows, first = [prows[i] for i in sel], [xs[i] for i in sel]
+ by, xs0, tm = _evaluate(recipe, critic, params, seeds, reps, workers, first=first)
+ timing['render_s'] += tm['render_s']
+ timing['score_s'] += tm['score_s']
+ timing['rendered'] += tm['rendered']
+ timing['scored'] += tm['scored']
+ if prows is None: # measurements for diverse picks and the report (cheap)
+ t0 = time.time()
+ prows = pre.score(xs0, polish=False) if pre is not None else [{'desc': describe(x)} if x is not None else {}
+ for x in xs0]
+ timing['pre_s'] += time.time() - t0
+ rows = []
+ for j in range(len(U)):
sc = by.get(j, {'total': -np.inf, 'passes': False, 'summary': 'render failed or silent'})
- history.append({'gen': g, 'u': U[j], 'params': params[j], 'seed': seeds[j], 'score': sc})
+ h = {'gen': g, 'u': U[j], 'params': params[j], 'seed': seeds[j], 'score': sc}
+ pr = prows[j] if j < len(prows) else {}
+ if pr.get('desc') is not None:
+ h['desc'] = pr['desc']
+ if 'pre' in pr and pre is not None and pre.n:
+ h['pre'] = {k: pr.get(k) for k in ('pre', 'tone_db', 'tone_ratio', 'env_pen')}
+ h['pre']['art'] = pr['art']['flags']
+ if 'polish' in pr:
+ h['pre']['polish'] = pr['polish']
+ rows.append(h)
+ history += rows
ranked = sorted([h for h in history if np.isfinite(h['score']['total'])], key=lambda h: -h['score']['total'])
elites = ranked[:elite]
+ fin = [h['score']['total'] for h in rows if np.isfinite(h['score']['total'])]
+ gs = {'gen': g + 1, 'best': round(float(elites[0]['score']['total']), 3) if elites else None,
+ 'gen_best': round(max(fin), 3) if fin else None,
+ 'gen_median': round(float(np.median(fin)), 3) if fin else None,
+ 'passes': sum(1 for h in rows if h['score'].get('passes')), 'scored': len(rows),
+ 'rendered': (K if use_pre else len(rows)), 'secs': round(time.time() - t_start, 1)}
+ if strategy != 'cma':
+ gs['sigma'] = round(sig, 4)
+ if elites and 'tone_db' in elites[0]['score']:
+ gs['best_tone_db'] = elites[0]['score']['tone_db']
+ if strategy == 'cma':
+ fu = [h['u'] for h in rows if np.isfinite(h['score']['total'])]
+ ft = [h['score']['total'] for h in rows if np.isfinite(h['score']['total'])]
+ if fu:
+ cma.tell(fu, ft)
+ gs['sigma'] = round(cma.sigma, 4)
+ if elites and elites[0]['score']['total'] > best_seen + 1e-3:
+ best_seen, stale = elites[0]['score']['total'], 0
+ else:
+ stale += 1
+ if g < gens - 1 and (cma.sigma < sigma_stop or stale >= patience):
+ restarts += 1
+ tried.append(cma.m.copy())
+ far = [h for h in ranked[:24] if min(np.linalg.norm(h['u'] - t) for t in tried) / math.sqrt(d) > 0.2]
+ m0 = far[0]['u'] if far and restarts % 2 == 1 else R.uniform(0, 1, d)
+ cma.reset(m0, co['sigma0'])
+ stale = 0
+ gs['restart'] = 'elite' if (far and restarts % 2 == 1) else 'random'
+ gstats.append(gs)
if log and elites:
b = elites[0]['score']
+ extra = f" sigma {gs['sigma']:.3f}" if 'sigma' in gs else ''
+ extra += f" (restart from {gs['restart']})" if gs.get('restart') else ''
+ extra += f' tone {b["tone_db"]:.1f} dB' if b.get('tone_db') is not None else ''
log(f"gen {g + 1}/{gens}: best total {b['total']:.2f} ({'PASS' if b['passes'] else 'no pass'}) "
- f"z {b.get('match_z')} p_real {b.get('p_real')} typ {b.get('ref_typ_z')} pen {b.get('feat_pen')}")
+ f"z {b.get('match_z')} p_real {b.get('p_real')} typ {b.get('ref_typ_z')} pen {b.get('feat_pen')}"
+ + extra + (f' [{K} rendered, {pop} scored]' if use_pre else ''))
ranked = sorted([h for h in history if np.isfinite(h['score']['total'])], key=lambda h: -h['score']['total'])
- return {'best': ranked[:max(elite, 8)], 'history': history, 'generations': gens}
+ timing['total_s'] = round(time.time() - t_start, 2)
+ for k in ('render_s', 'pre_s', 'score_s'):
+ timing[k] = round(timing[k], 2)
+ timing['scored_per_min'] = round(len(history) / max(timing['total_s'], 1e-9) * 60, 1)
+ timing['rendered_per_min'] = round(timing['rendered'] / max(timing['total_s'], 1e-9) * 60, 1)
+ return {'best': ranked[:max(elite, 8)], 'history': history, 'generations': gens, 'gens': gstats,
+ 'timing': timing, 'strategy': strategy, 'prescreen': K if use_pre else 0, 'restarts': restarts,
+ 'full_space': full}
-def robust(recipe, critic, params, seeds=4, seed0=5000):
+def robust(recipe, critic, params, seeds=4, seed0=5000, workers=None):
"""re-render `params` on `seeds` new noise seeds (seed0, seed0+1, ...) and score them: {mean, min, max, sd,
pass_rate, n, totals}. A sound that only scores well on one seed is a lucky render, not a good recipe."""
ss = list(range(seed0, seed0 + seeds)) if isinstance(seeds, int) else list(seeds)
- xs = [render(recipe, params, s) for s in ss]
+ xs = render_many(recipe, [(params, s) for s in ss], workers)
live = [x for x in xs if x is not None]
rows = critic.score(live) if live else []
fin = [float(r['total']) for r in rows if np.isfinite(r['total'])]
- return {'n': len(ss), 'seeds': ss, 'totals': [round(t, 3) for t in fin], 'failed': len(ss) - len(fin),
- 'mean': round(float(np.mean(fin)), 3) if fin else None, 'min': round(float(min(fin)), 3) if fin else None,
- 'max': round(float(max(fin)), 3) if fin else None, 'sd': round(float(np.std(fin)), 3) if fin else None,
- 'pass_rate': round(sum(1 for r in rows if r.get('passes')) / len(ss), 3)}
+ out = {'n': len(ss), 'seeds': ss, 'totals': [round(t, 3) for t in fin], 'failed': len(ss) - len(fin),
+ 'mean': round(float(np.mean(fin)), 3) if fin else None, 'min': round(float(min(fin)), 3) if fin else None,
+ 'max': round(float(max(fin)), 3) if fin else None, 'sd': round(float(np.std(fin)), 3) if fin else None,
+ 'pass_rate': round(sum(1 for r in rows if r.get('passes')) / len(ss), 3)}
+ tdb = [r['tone_db'] for r in rows if r.get('tone_db') is not None]
+ if tdb:
+ out['tone_db_mean'] = round(float(np.mean(tdb)), 2)
+ return out
+
+
+def diverse(results, recipe, k=4, min_dist=0.08, min_sound=2.0):
+ """pick up to k strong results, passing first, whose parameter vectors differ (normalised distance >= min_dist)
+ and, when the results carry sound descriptors ('desc', from optimize), whose sounds differ too (sound_dist >=
+ min_sound dB-like units: tonal balance + envelope). If that leaves fewer than k, the rest are filled with the
+ parameter rule alone (the v1 rule)."""
+ order = sorted(results, key=lambda h: (not h['score'].get('passes'), -h['score']['total']))
+ order = [h for h in order if np.isfinite(h['score']['total'])] or order
+
+ def pdist(a, b):
+ return np.linalg.norm(a['u'] - b['u']) / math.sqrt(len(a['u']))
+ def ok(h, out, sound):
+ for o in out:
+ if o is h or pdist(h, o) < min_dist:
+ return False
+ if sound and 'desc' in h and 'desc' in o and sound_dist(h['desc'], o['desc']) < min_sound:
+ return False
+ return True
-def diverse(results, recipe, k=4, min_dist=0.08):
- """pick up to k strong results whose parameter vectors differ (normalised distance >= min_dist), passing first"""
out = []
- for h in sorted(results, key=lambda h: (not h['score'].get('passes'), -h['score']['total'])):
- if all(np.linalg.norm(h['u'] - o['u']) / math.sqrt(len(h['u'])) >= min_dist for o in out):
- out.append(h)
- if len(out) == k:
- break
- return out
+ for sound in ((True, False) if min_sound else (False,)):
+ for h in order:
+ if len(out) == k:
+ break
+ if all(h is not o for o in out) and ok(h, out, sound):
+ out.append(h)
+ return sorted(out, key=lambda h: (not h['score'].get('passes'), -h['score']['total']))
+
+
+# ---------------------------------------------------------------------------------------------- seed bank
+def _ctx(target, refs):
+ return (target or '') + ' | ' + ','.join(sorted(os.path.basename(str(r)) for r in (refs or [])))
+
+
+def _round_params(p):
+ return {k: (float(f'{v:.6g}') if isinstance(v, float) else v) for k, v in p.items()}
+
+
+def seeds_path(recipe, folder=None):
+ name = recipe if isinstance(recipe, str) else getattr(recipe, '__name__', 'recipe').rsplit('.', 1)[-1]
+ return os.path.join(folder or SEEDS_DIR, f'{name}.json')
+
+
+def load_seeds(recipe, target=None, refs=None, n=6, folder=None):
+ """good parameter sets found so far for a recipe (earshot/forge/seeds/<recipe>.json): up to n param dicts, the
+ ones found with the same target + refs first, then the rest by score. [] when there is no bank."""
+ path = seeds_path(recipe, folder)
+ if not os.path.exists(path):
+ return []
+ with open(path, encoding='utf-8') as fh:
+ bank = json.load(fh)
+ ctx = _ctx(target, refs)
+ ents = sorted(bank.get('entries', []), key=lambda e: (e.get('context') != ctx, -float(e.get('rank_score', -1e9))))
+ return [dict(e['params']) for e in ents[:n]]
+
+
+def save_seeds(recipe, picks, target=None, refs=None, source='', folder=None, per_context=6, cap=12):
+ """merge picks ({params, score[, robust]} dicts) into the recipe's seed bank; ranked by the robust mean when there
+ is one, else the total. Keeps it small: per_context entries per target + refs, cap in all (the newest contexts
+ win), near-duplicate parameter sets merged. Atomic write. Returns the bank dict."""
+ path = seeds_path(recipe, folder)
+ bank = {'recipe': os.path.basename(path)[:-5], 'entries': []}
+ if os.path.exists(path):
+ with open(path, encoding='utf-8') as fh:
+ bank = json.load(fh)
+ ctx = _ctx(target, refs)
+ new = []
+ for p in picks:
+ sc = p.get('score', {})
+ rb = p.get('robust') or {}
+ rank = rb.get('mean') if rb.get('mean') is not None else sc.get('total')
+ if rank is None or not np.isfinite(rank):
+ continue
+ e = {'params': _round_params(p['params']), 'rank_score': round(float(rank), 3), 'context': ctx,
+ 'total': sc.get('total'), 'tone_db': sc.get('tone_db'), 'passes': bool(sc.get('passes')),
+ 'source': source}
+ if rb:
+ e['robust_mean'], e['robust_pass_rate'] = rb.get('mean'), rb.get('pass_rate')
+ new.append(e)
+ ents = new + [e for e in bank.get('entries', [])]
+ keep, seen_ctx = [], []
+ for e in ents:
+ if e['context'] not in seen_ctx:
+ seen_ctx.append(e['context'])
+ for c in seen_ctx:
+ mine = sorted([e for e in ents if e['context'] == c], key=lambda e: -float(e['rank_score']))
+ picked = []
+ for e in mine:
+ if all(e['params'] != q['params'] for q in picked):
+ picked.append(e)
+ if len(picked) == per_context:
+ break
+ keep += picked
+ bank['entries'] = keep[:cap]
+ os.makedirs(os.path.dirname(path), exist_ok=True)
+ tmp = path + f'.tmp{os.getpid()}'
+ with open(tmp, 'w', encoding='utf-8') as fh:
+ json.dump(bank, fh, indent=1)
+ os.replace(tmp, path)
+ return bank
diff --git a/earshot/forge/seeds/click.json b/earshot/forge/seeds/click.json
new file mode 100644
index 0000000..059deb7
--- /dev/null
+++ b/earshot/forge/seeds/click.json
@@ -0,0 +1,146 @@
+{
+ "recipe": "click",
+ "entries": [
+ {
+ "params": {
+ "thump_amp": 0.11751,
+ "thump_f": 250.0,
+ "thump_w": 0.00460395,
+ "pre": 0.00188092,
+ "snap_w": 0.00012,
+ "snap_bright": 0.452066,
+ "direct": 0.426035,
+ "main_f": 3579.09,
+ "main_decay": 0.000473332,
+ "main_amp": 0.697358,
+ "tick_lvl": 0.0204845,
+ "tick_dt": 0.00279366,
+ "radiate_hz": 131.899,
+ "shell_f": 3167.95,
+ "shell_n": 9,
+ "shell_decay": 0.00249802,
+ "grain": 0.0622303,
+ "body_f": 1020.26,
+ "body_decay": 0.00193485,
+ "body_amp": 0.26458,
+ "chatter": 1,
+ "chatter_dt": 0.000627487,
+ "bottom_dt": 0.0015,
+ "bottom_lvl": 0.25881,
+ "rel_dt": 0.0331076,
+ "rel_lvl": 0.0476638,
+ "tail_lvl": 0.0311158,
+ "tail_tau": 0.0146674,
+ "tail_f": 3385.75,
+ "tail_bw": 2.34319,
+ "room_wet": 0.123184,
+ "room_damp": 0.127211,
+ "rt60": 0.0667033,
+ "floor_db": -58.0298,
+ "hiss_tilt": 0.860933,
+ "rumble_db": -72.2234,
+ "clip_len": 0.0936147
+ },
+ "rank_score": 5.741,
+ "context": "a soft premium UI button click | ui_click_01.wav,ui_click_02.wav,ui_click_03.wav,ui_click_04.wav,ui_toggle_01.wav,ui_toggle_02.wav",
+ "total": 5.741,
+ "tone_db": null,
+ "passes": false,
+ "source": "v1 forge-test final (demo/forge-test/click/final)"
+ },
+ {
+ "params": {
+ "thump_amp": 0.12,
+ "thump_f": 250.0,
+ "thump_w": 0.005,
+ "pre": 0.00157604,
+ "snap_w": 0.00011792,
+ "snap_bright": 0.499291,
+ "direct": 0.472607,
+ "main_f": 3232.17,
+ "main_decay": 0.000422276,
+ "main_amp": 0.74094,
+ "tick_lvl": 0.0302383,
+ "tick_dt": 0.00258793,
+ "radiate_hz": 119.862,
+ "shell_f": 3507.72,
+ "shell_n": 8,
+ "shell_decay": 0.00241072,
+ "grain": 0.05,
+ "body_f": 871.607,
+ "body_decay": 0.00175332,
+ "body_amp": 0.177341,
+ "chatter": 2,
+ "chatter_dt": 0.000688231,
+ "bottom_dt": 0.0015,
+ "bottom_lvl": 0.21142,
+ "rel_dt": 0.031779,
+ "rel_lvl": 0.0460895,
+ "tail_lvl": 0.0116713,
+ "tail_tau": 0.0134047,
+ "tail_f": 4401.91,
+ "tail_bw": 2.60375,
+ "room_wet": 0.063841,
+ "room_damp": 0.179332,
+ "rt60": 0.081235,
+ "floor_db": -60.1345,
+ "hiss_tilt": 1.0,
+ "rumble_db": -72.8138,
+ "clip_len": 0.078428
+ },
+ "rank_score": 5.671,
+ "context": "a soft premium UI button click | ui_click_01.wav,ui_click_02.wav,ui_click_03.wav,ui_click_04.wav,ui_toggle_01.wav,ui_toggle_02.wav",
+ "total": 5.671,
+ "tone_db": null,
+ "passes": true,
+ "source": "v1 forge-test final (demo/forge-test/click/final)"
+ },
+ {
+ "params": {
+ "thump_amp": 0.0430346,
+ "thump_f": 289.285,
+ "thump_w": 0.005,
+ "pre": 0.00233468,
+ "snap_w": 8.69414e-05,
+ "snap_bright": 0.467189,
+ "direct": 0.564955,
+ "main_f": 2809.51,
+ "main_decay": 0.000554604,
+ "main_amp": 0.438862,
+ "tick_lvl": 0.00897003,
+ "tick_dt": 0.00274409,
+ "radiate_hz": 60.6098,
+ "shell_f": 4087.75,
+ "shell_n": 8,
+ "shell_decay": 0.00201662,
+ "grain": 0.145923,
+ "body_f": 828.632,
+ "body_decay": 0.00281656,
+ "body_amp": 0.179248,
+ "chatter": 0,
+ "chatter_dt": 0.000608129,
+ "bottom_dt": 0.0015,
+ "bottom_lvl": 0.289764,
+ "rel_dt": 0.036,
+ "rel_lvl": 0.0401836,
+ "tail_lvl": 0.0435149,
+ "tail_tau": 0.00789143,
+ "tail_f": 3157.96,
+ "tail_bw": 2.02053,
+ "room_wet": 0.0387742,
+ "room_damp": 0.144004,
+ "rt60": 0.127354,
+ "floor_db": -59.8835,
+ "hiss_tilt": 0.589251,
+ "rumble_db": -80.0,
+ "clip_len": 0.0790152
+ },
+ "rank_score": 5.453,
+ "context": "a soft premium UI button click | ui_click_01.wav,ui_click_02.wav,ui_click_03.wav,ui_click_04.wav,ui_toggle_01.wav,ui_toggle_02.wav",
+ "total": 5.453,
+ "tone_db": null,
+ "passes": false,
+ "source": "v1 forge-test final (demo/forge-test/click/final)"
+ }
+ ]
+}
\ No newline at end of file
diff --git a/earshot/forge/seeds/impact.json b/earshot/forge/seeds/impact.json
new file mode 100644
index 0000000..c45a102
--- /dev/null
+++ b/earshot/forge/seeds/impact.json
@@ -0,0 +1,167 @@
+{
+ "recipe": "impact",
+ "entries": [
+ {
+ "params": {
+ "pre_len": 0.787543,
+ "pre_level": 0.133024,
+ "pre_bright": 600.0,
+ "pre_low": 0.577469,
+ "riser_level": 0.0,
+ "riser_f": 3063.0,
+ "crack_level": 0.719992,
+ "crack_ms": 1.9069,
+ "crack_f": 1133.17,
+ "contact_ms": 0.630756,
+ "crackle": 1.03506,
+ "debris_ms": 66.9317,
+ "thump_f0": 141.859,
+ "thump_f1": 30.1634,
+ "drop_tau": 0.0345376,
+ "thump_tau": 0.208735,
+ "modes_level": 0.263301,
+ "shell_level": 0.203514,
+ "shell_lo": 121.74,
+ "shell_span": 3.56957,
+ "shell_tau": 0.350685,
+ "sub_f": 60.131,
+ "sub_tau": 1.05181,
+ "sub_level": 0.0197525,
+ "rumble_level": 0.0589769,
+ "rumble_cut": 103.876,
+ "rumble_tau": 2.3428,
+ "sweep_level": 0.11233,
+ "sweep_hi": 365.866,
+ "sweep_tau": 0.569212,
+ "echo_level": 0.423833,
+ "echo_delay": 0.119247,
+ "comb_mix": 0.167738,
+ "comb_d0": 0.245297,
+ "comb_ratio": 3.23043,
+ "comb_fb": 0.752118,
+ "hall_wet": 0.790224,
+ "hall_rt": 3.04516,
+ "hall_damp": 0.288663,
+ "predelay": 0.0,
+ "drive": 0.352919,
+ "sat_tone": 8002.76,
+ "width": 0.336321,
+ "floor_db": -88.6703
+ },
+ "rank_score": 7.228,
+ "context": "a deep cinematic boom impact | impact_01.wav,impact_02.wav,impact_03.wav",
+ "total": 7.228,
+ "tone_db": null,
+ "passes": true,
+ "source": "v1 forge-test final (demo/forge-test/impact/final)"
+ },
+ {
+ "params": {
+ "pre_len": 0.816053,
+ "pre_level": 0.138728,
+ "pre_bright": 675.23,
+ "pre_low": 0.525659,
+ "riser_level": 0.00148853,
+ "riser_f": 3296.18,
+ "crack_level": 0.764693,
+ "crack_ms": 3.0,
+ "crack_f": 691.473,
+ "contact_ms": 0.316741,
+ "crackle": 0.94546,
+ "debris_ms": 60.0,
+ "thump_f0": 188.572,
+ "thump_f1": 33.3514,
+ "drop_tau": 0.0196134,
+ "thump_tau": 0.293252,
+ "modes_level": 0.106729,
+ "shell_level": 0.201194,
+ "shell_lo": 127.381,
+ "shell_span": 5.73038,
+ "shell_tau": 0.627228,
+ "sub_f": 58.9831,
+ "sub_tau": 0.654219,
+ "sub_level": 0.11982,
+ "rumble_level": 0.06019,
+ "rumble_cut": 194.151,
+ "rumble_tau": 1.90249,
+ "sweep_level": 0.179511,
+ "sweep_hi": 616.857,
+ "sweep_tau": 0.510043,
+ "echo_level": 0.283771,
+ "echo_delay": 0.137539,
+ "comb_mix": 0.0,
+ "comb_d0": 0.306828,
+ "comb_ratio": 3.07961,
+ "comb_fb": 0.495804,
+ "hall_wet": 0.909031,
+ "hall_rt": 4.45399,
+ "hall_damp": 0.261761,
+ "predelay": 0.00241012,
+ "drive": 0.269575,
+ "sat_tone": 5711.94,
+ "width": 0.395563,
+ "floor_db": -85.3966
+ },
+ "rank_score": 7.193,
+ "context": "a deep cinematic boom impact | impact_01.wav,impact_02.wav,impact_03.wav",
+ "total": 7.193,
+ "tone_db": null,
+ "passes": true,
+ "source": "v1 forge-test final (demo/forge-test/impact/final)"
+ },
+ {
+ "params": {
+ "pre_len": 1.04073,
+ "pre_level": 0.0959787,
+ "pre_bright": 954.889,
+ "pre_low": 0.358539,
+ "riser_level": 0.0109828,
+ "riser_f": 4115.0,
+ "crack_level": 0.8,
+ "crack_ms": 2.46133,
+ "crack_f": 1281.52,
+ "contact_ms": 0.151787,
+ "crackle": 0.565127,
+ "debris_ms": 60.0,
+ "thump_f0": 171.279,
+ "thump_f1": 29.8399,
+ "drop_tau": 0.126255,
+ "thump_tau": 0.215735,
+ "modes_level": 0.0756361,
+ "shell_level": 0.456485,
+ "shell_lo": 115.259,
+ "shell_span": 3.96542,
+ "shell_tau": 0.560587,
+ "sub_f": 62.0,
+ "sub_tau": 0.909703,
+ "sub_level": 0.135047,
+ "rumble_level": 0.0708323,
+ "rumble_cut": 282.658,
+ "rumble_tau": 2.08079,
+ "sweep_level": 0.108992,
+ "sweep_hi": 801.304,
+ "sweep_tau": 0.566856,
+ "echo_level": 0.330159,
+ "echo_delay": 0.349633,
+ "comb_mix": 0.0238023,
+ "comb_d0": 0.378327,
+ "comb_ratio": 5.40663,
+ "comb_fb": 0.470344,
+ "hall_wet": 1.13233,
+ "hall_rt": 2.31042,
+ "hall_damp": 0.0580355,
+ "predelay": 0.0162008,
+ "drive": 0.251733,
+ "sat_tone": 9519.89,
+ "width": 0.0947279,
+ "floor_db": -85.9915
+ },
+ "rank_score": 7.094,
+ "context": "a deep cinematic boom impact | impact_01.wav,impact_02.wav,impact_03.wav",
+ "total": 7.094,
+ "tone_db": null,
+ "passes": true,
+ "source": "v1 forge-test final (demo/forge-test/impact/final)"
+ }
+ ]
+}
\ No newline at end of file
diff --git a/earshot/forge/seeds/keypress.json b/earshot/forge/seeds/keypress.json
new file mode 100644
index 0000000..80fd085
--- /dev/null
+++ b/earshot/forge/seeds/keypress.json
@@ -0,0 +1,173 @@
+{
+ "recipe": "keypress",
+ "entries": [
+ {
+ "params": {
+ "click_tc": 7.62977e-05,
+ "click_tau": 0.000400034,
+ "click_lvl": 0.195527,
+ "click_lo": 7323.18,
+ "click_tilt": -1.25571,
+ "crack": 0.395643,
+ "snap": 0.0,
+ "cap_f": 1489.18,
+ "cap_tau": 0.00399449,
+ "cap_noise": 1.0,
+ "cap_n": 7,
+ "cap_cl": 0.357156,
+ "cap_fast": 0.000995191,
+ "cap_res": 0.749554,
+ "mid_f": 205.793,
+ "mid_tau": 0.00918634,
+ "mid_lvl": 0.0902506,
+ "thock_f": 86.8567,
+ "thock_tau": 0.0549253,
+ "thock_lvl": 0.226067,
+ "thock_rise": 0.00834742,
+ "case_f": 70.0,
+ "case_lvl": 0.581134,
+ "case_dec": 0.736866,
+ "up_delay": 0.124575,
+ "up_lvl": 0.0969741,
+ "up_pitch": 1.04402,
+ "tick_lvl": 0.581597,
+ "tick_dt": 0.0111911,
+ "pre_len": 0.0128195,
+ "pre_lvl": 0.0207414,
+ "ping_f": 9083.43,
+ "ping_tau": 0.00578368,
+ "ping_lvl": 0.122274,
+ "room_wet": 0.261726,
+ "rt60": 0.113685,
+ "room_size": 0.212985,
+ "room_damp": 0.95,
+ "room_model": "sparse",
+ "floor_db": -55.0,
+ "rumble": 0.0177183,
+ "clip": 0.38353,
+ "hpf": 44.5337,
+ "key": 999
+ },
+ "rank_score": 6.736,
+ "context": "a single deep thocky mechanical keyboard key press | kb_enter_01.wav,kb_space_01.wav,kb_thock_01.wav,kb_thock_02.wav,kb_thock_03.wav,kb_thock_04.wav",
+ "total": 6.749,
+ "tone_db": 5.6,
+ "passes": true,
+ "source": "v1 forge-test final (demo/forge-test/keypress/final)",
+ "robust_mean": 6.736,
+ "robust_pass_rate": 0.875
+ },
+ {
+ "params": {
+ "click_tc": 6.37017e-05,
+ "click_tau": 0.000335592,
+ "click_lvl": 0.320482,
+ "click_lo": 7388.63,
+ "click_tilt": -1.28867,
+ "crack": 0.323781,
+ "snap": 0.0378001,
+ "cap_f": 1233.34,
+ "cap_tau": 0.00425304,
+ "cap_noise": 0.985326,
+ "cap_n": 6,
+ "cap_cl": 0.373503,
+ "cap_fast": 0.000839102,
+ "cap_res": 0.845025,
+ "mid_f": 209.773,
+ "mid_tau": 0.00856942,
+ "mid_lvl": 0.142669,
+ "thock_f": 82.8638,
+ "thock_tau": 0.045,
+ "thock_lvl": 0.278363,
+ "thock_rise": 0.00678305,
+ "case_f": 70.0,
+ "case_lvl": 0.594141,
+ "case_dec": 0.761849,
+ "up_delay": 0.125,
+ "up_lvl": 0.112565,
+ "up_pitch": 1.07032,
+ "tick_lvl": 0.357141,
+ "tick_dt": 0.010767,
+ "pre_len": 0.0118728,
+ "pre_lvl": 0.0150143,
+ "ping_f": 8894.51,
+ "ping_tau": 0.00658933,
+ "ping_lvl": 0.10548,
+ "room_wet": 0.234312,
+ "rt60": 0.131531,
+ "room_size": 0.0952076,
+ "room_damp": 0.95,
+ "room_model": "sparse",
+ "floor_db": -58.529,
+ "rumble": 0.029323,
+ "clip": 0.359842,
+ "hpf": 46.0,
+ "key": 981
+ },
+ "rank_score": 6.732,
+ "context": "a single deep thocky mechanical keyboard key press | kb_enter_01.wav,kb_space_01.wav,kb_thock_01.wav,kb_thock_02.wav,kb_thock_03.wav,kb_thock_04.wav",
+ "total": 7.374,
+ "tone_db": 5.54,
+ "passes": true,
+ "source": "v1 forge-test final (demo/forge-test/keypress/final)",
+ "robust_mean": 6.732,
+ "robust_pass_rate": 0.75
+ },
+ {
+ "params": {
+ "click_tc": 7.71749e-05,
+ "click_tau": 0.000308083,
+ "click_lvl": 0.207913,
+ "click_lo": 7178.75,
+ "click_tilt": -1.68287,
+ "crack": 0.418696,
+ "snap": 0.0,
+ "cap_f": 1229.2,
+ "cap_tau": 0.0045,
+ "cap_noise": 0.952392,
+ "cap_n": 6,
+ "cap_cl": 0.333026,
+ "cap_fast": 0.000957235,
+ "cap_res": 0.833536,
+ "mid_f": 236.447,
+ "mid_tau": 0.00690859,
+ "mid_lvl": 0.129485,
+ "thock_f": 81.8564,
+ "thock_tau": 0.0454297,
+ "thock_lvl": 0.293337,
+ "thock_rise": 0.00683346,
+ "case_f": 73.5898,
+ "case_lvl": 0.558731,
+ "case_dec": 0.687316,
+ "up_delay": 0.122283,
+ "up_lvl": 0.0811195,
+ "up_pitch": 0.929842,
+ "tick_lvl": 0.333061,
+ "tick_dt": 0.0114672,
+ "pre_len": 0.0134625,
+ "pre_lvl": 0.0214919,
+ "ping_f": 9183.34,
+ "ping_tau": 0.00701581,
+ "ping_lvl": 0.0966674,
+ "room_wet": 0.228344,
+ "rt60": 0.139951,
+ "room_size": 0.0709654,
+ "room_damp": 0.936431,
+ "room_model": "sparse",
+ "floor_db": -63.4449,
+ "rumble": 0.084591,
+ "clip": 0.368763,
+ "hpf": 44.7958,
+ "key": 999
+ },
+ "rank_score": 6.156,
+ "context": "a single deep thocky mechanical keyboard key press | kb_enter_01.wav,kb_space_01.wav,kb_thock_01.wav,kb_thock_02.wav,kb_thock_03.wav,kb_thock_04.wav",
+ "total": 6.787,
+ "tone_db": 5.6,
+ "passes": true,
+ "source": "v1 forge-test final (demo/forge-test/keypress/final)",
+ "robust_mean": 6.156,
+ "robust_pass_rate": 0.625
+ }
+ ]
+}
\ No newline at end of file
diff --git a/earshot/forge/seeds/pop.json b/earshot/forge/seeds/pop.json
new file mode 100644
index 0000000..f20d04a
--- /dev/null
+++ b/earshot/forge/seeds/pop.json
@@ -0,0 +1,175 @@
+{
+ "recipe": "pop",
+ "entries": [
+ {
+ "params": {
+ "f0": 487.564,
+ "q": 22.8501,
+ "rise": 1.06344,
+ "rise_tau": 0.0344421,
+ "att": 0.006,
+ "h2": 0.0738033,
+ "wobble": 0.0123015,
+ "breath_amp": 0.0335639,
+ "breath_tau": 0.00120778,
+ "f2_ratio": 2.85139,
+ "f2_amp": 0.520685,
+ "f2_q": 23.3773,
+ "f2_follow": 0.941618,
+ "sub_amp": 0.142645,
+ "sub_ratio": 1.82928,
+ "sub_delay": 0.00487339,
+ "click_amp": 0.159337,
+ "click_w": 0.000169091,
+ "tear": 0.000409203,
+ "jerks": 1,
+ "click_lp": 7345.44,
+ "shred_amp": 0.194455,
+ "shred_tau": 0.000683151,
+ "shred_hp": 4767.04,
+ "hiss_amp": 0.0168237,
+ "hiss_f": 3261.85,
+ "hiss_q": 2.22709,
+ "hiss_tau": 0.00151189,
+ "puff_amp": 0.166349,
+ "puff_f": 120.0,
+ "ring_delay": 0.0,
+ "rt60": 0.242573,
+ "wet": 0.39907,
+ "damp": 0.72231,
+ "size": 0.154421,
+ "air_db": -4.90373,
+ "floor_db": -70.9847,
+ "swell_amp": 0.0157551,
+ "swell_t": 0.0101016,
+ "pre": 0.0336202,
+ "hp": 45.6866,
+ "lp": 11759.8,
+ "clip": 0.143547,
+ "cut_fade": 0.00935593,
+ "cut_db": -48.4792
+ },
+ "rank_score": 6.798,
+ "context": "a small bubble pop | ui_pop_01.wav,ui_pop_02.wav,ui_pop_03.wav,ui_pop_04.wav",
+ "total": 7.012,
+ "tone_db": 9.41,
+ "passes": true,
+ "source": "v1 forge-test final (demo/forge-test/pop/final)",
+ "robust_mean": 6.798,
+ "robust_pass_rate": 1.0
+ },
+ {
+ "params": {
+ "f0": 440.159,
+ "q": 22.1207,
+ "rise": 0.991892,
+ "rise_tau": 0.0159722,
+ "att": 0.000357098,
+ "h2": 0.00847939,
+ "wobble": 0.0112906,
+ "breath_amp": 0.0,
+ "breath_tau": 0.00848874,
+ "f2_ratio": 3.82514,
+ "f2_amp": 0.8,
+ "f2_q": 25.0,
+ "f2_follow": 0.948223,
+ "sub_amp": 0.00328062,
+ "sub_ratio": 2.08913,
+ "sub_delay": 0.0111736,
+ "click_amp": 0.685048,
+ "click_w": 7.72837e-05,
+ "tear": 0.000171419,
+ "jerks": 1,
+ "click_lp": 4881.31,
+ "shred_amp": 0.200779,
+ "shred_tau": 0.00153019,
+ "shred_hp": 5000.0,
+ "hiss_amp": 0.0659931,
+ "hiss_f": 2418.88,
+ "hiss_q": 2.86992,
+ "hiss_tau": 0.00113073,
+ "puff_amp": 0.0,
+ "puff_f": 121.063,
+ "ring_delay": 0.00105542,
+ "rt60": 0.249181,
+ "wet": 0.291252,
+ "damp": 0.490701,
+ "size": 0.05,
+ "air_db": -0.476997,
+ "floor_db": -64.4367,
+ "swell_amp": 0.160452,
+ "swell_t": 0.00252584,
+ "pre": 0.0269276,
+ "hp": 92.6866,
+ "lp": 10181.4,
+ "clip": 0.274425,
+ "cut_fade": 0.0153059
+ },
+ "rank_score": 6.759,
+ "context": "a small bubble pop | ui_pop_01.wav,ui_pop_02.wav,ui_pop_03.wav,ui_pop_04.wav",
+ "total": 6.802,
+ "tone_db": 9.37,
+ "passes": true,
+ "source": "v1 forge-test final (demo/forge-test/pop/final)",
+ "robust_mean": 6.759,
+ "robust_pass_rate": 0.875
+ },
+ {
+ "params": {
+ "f0": 517.457,
+ "q": 22.7738,
+ "rise": 1.01333,
+ "rise_tau": 0.0334035,
+ "att": 0.006,
+ "h2": 0.0197352,
+ "wobble": 0.00483019,
+ "breath_amp": 0.0146075,
+ "breath_tau": 0.0014566,
+ "f2_ratio": 2.691,
+ "f2_amp": 0.38407,
+ "f2_q": 25.0,
+ "f2_follow": 0.892632,
+ "sub_amp": 0.365912,
+ "sub_ratio": 1.80786,
+ "sub_delay": 0.00735194,
+ "click_amp": 0.145113,
+ "click_w": 0.000230256,
+ "tear": 0.000276229,
+ "jerks": 1,
+ "click_lp": 7574.89,
+ "shred_amp": 0.193232,
+ "shred_tau": 0.00109936,
+ "shred_hp": 3736.06,
+ "hiss_amp": 0.0517719,
+ "hiss_f": 2534.9,
+ "hiss_q": 1.8692,
+ "hiss_tau": 0.00136812,
+ "puff_amp": 0.169544,
+ "puff_f": 121.572,
+ "ring_delay": 9.98099e-05,
+ "rt60": 0.214609,
+ "wet": 0.367916,
+ "damp": 0.878127,
+ "size": 0.146303,
+ "air_db": -3.84319,
+ "floor_db": -70.5919,
+ "swell_amp": 0.0736066,
+ "swell_t": 0.00496391,
+ "pre": 0.0336165,
+ "hp": 45.7743,
+ "lp": 12820.4,
+ "clip": 0.144714,
+ "cut_fade": 0.00852325,
+ "cut_db": -47.8559
+ },
+ "rank_score": 6.729,
+ "context": "a small bubble pop | ui_pop_01.wav,ui_pop_02.wav,ui_pop_03.wav,ui_pop_04.wav",
+ "total": 6.96,
+ "tone_db": 7.33,
+ "passes": true,
+ "source": "v1 forge-test final (demo/forge-test/pop/final)",
+ "robust_mean": 6.729,
+ "robust_pass_rate": 1.0
+ }
+ ]
+}
\ No newline at end of file
diff --git a/earshot/forge/seeds/whoosh.json b/earshot/forge/seeds/whoosh.json
new file mode 100644
index 0000000..994b87d
--- /dev/null
+++ b/earshot/forge/seeds/whoosh.json
@@ -0,0 +1,161 @@
+{
+ "recipe": "whoosh",
+ "entries": [
+ {
+ "params": {
+ "tc": 0.0706346,
+ "s_rise": 0.0380857,
+ "s_fall": 0.0578346,
+ "beta": 1.49847,
+ "prox_w": 0.0243735,
+ "f_peak": 398.391,
+ "f_track": 0.108498,
+ "q_shed": 7.05767,
+ "steep": 0.23086,
+ "a_pow": 2.90371,
+ "doppler": 0.00143736,
+ "wake_db": -14.0966,
+ "wake_tau": 0.0532322,
+ "wake_q": 4.45435,
+ "wake_dark": 0.613911,
+ "wake_hiss_db": -16.0,
+ "hiss_db": -19.345,
+ "hiss_hp": 1671.88,
+ "hiss_f": 9908.19,
+ "hiss_pow": 2.57151,
+ "push_db": -4.56985,
+ "push_w": 0.0251241,
+ "push_f": 125.898,
+ "buffet_db": -9.18542,
+ "buffet_f": 95.0096,
+ "buffet_delay": 0.0219106,
+ "buffet_tau": 0.0262978,
+ "flut_depth": 0.0178753,
+ "flut_f": 30.2817,
+ "whip_db": -16.8109,
+ "whip_q": 22.6043,
+ "whip_ratio": 1.75668,
+ "pan_w": 0.102279,
+ "width": 0.0730709,
+ "room_wet": 0.158579,
+ "rt60": 0.138233,
+ "air_db": -12.0,
+ "floor_db": -75.3278,
+ "cut_db": -46.4213,
+ "cut_fade": 0.0281062
+ },
+ "rank_score": 8.017,
+ "context": "a fast crisp air whoosh | whoosh_fast_01.wav,whoosh_fast_02.wav,whoosh_fast_03.wav,whoosh_fast_04.wav",
+ "total": 8.107,
+ "tone_db": 5.38,
+ "passes": true,
+ "source": "v1 forge-test final (demo/forge-test/whoosh/final)",
+ "robust_mean": 8.017,
+ "robust_pass_rate": 1.0
+ },
+ {
+ "params": {
+ "tc": 0.129107,
+ "s_rise": 0.0427788,
+ "s_fall": 0.0265824,
+ "beta": 1.54686,
+ "prox_w": 0.0339457,
+ "f_peak": 359.809,
+ "f_track": 0.0,
+ "q_shed": 12.0,
+ "steep": 0.200505,
+ "a_pow": 3.06338,
+ "doppler": 0.00366254,
+ "wake_db": -10.2002,
+ "wake_tau": 0.0607404,
+ "wake_q": 4.77622,
+ "wake_dark": 0.555271,
+ "wake_hiss_db": -23.0435,
+ "hiss_db": -25.0394,
+ "hiss_hp": 1657.34,
+ "hiss_f": 14000.0,
+ "hiss_pow": 3.58732,
+ "push_db": -14.4953,
+ "push_w": 0.0101188,
+ "push_f": 139.219,
+ "buffet_db": -2.94905,
+ "buffet_f": 63.2878,
+ "buffet_delay": 0.0398184,
+ "buffet_tau": 0.0200667,
+ "flut_depth": 0.144679,
+ "flut_f": 60.0,
+ "whip_db": -26.68,
+ "whip_q": 27.7504,
+ "whip_ratio": 2.65226,
+ "pan_w": 0.095567,
+ "width": 0.0897789,
+ "room_wet": 0.34712,
+ "rt60": 0.128766,
+ "air_db": -11.8299,
+ "floor_db": -72.1756,
+ "cut_db": -48.6671,
+ "cut_fade": 0.0440664
+ },
+ "rank_score": 7.821,
+ "context": "a fast crisp air whoosh | whoosh_fast_01.wav,whoosh_fast_02.wav,whoosh_fast_03.wav,whoosh_fast_04.wav",
+ "total": 8.019,
+ "tone_db": 4.57,
+ "passes": true,
+ "source": "v1 forge-test final (demo/forge-test/whoosh/final)",
+ "robust_mean": 7.821,
+ "robust_pass_rate": 1.0
+ },
+ {
+ "params": {
+ "tc": 0.0783005,
+ "s_rise": 0.0456235,
+ "s_fall": 0.0531854,
+ "beta": 1.34242,
+ "prox_w": 0.0265794,
+ "f_peak": 307.416,
+ "f_track": 0.0680297,
+ "q_shed": 8.39312,
+ "steep": 0.100729,
+ "a_pow": 2.81839,
+ "doppler": 0.0121328,
+ "wake_db": -14.0473,
+ "wake_tau": 0.0528149,
+ "wake_q": 4.06986,
+ "wake_dark": 0.569154,
+ "wake_hiss_db": -18.6524,
+ "hiss_db": -21.8907,
+ "hiss_hp": 1589.74,
+ "hiss_f": 9919.43,
+ "hiss_pow": 2.94198,
+ "push_db": -2.93574,
+ "push_w": 0.0322701,
+ "push_f": 115.677,
+ "buffet_db": -12.2581,
+ "buffet_f": 84.9928,
+ "buffet_delay": 0.0218329,
+ "buffet_tau": 0.0338947,
+ "flut_depth": 0.00669124,
+ "flut_f": 28.6802,
+ "whip_db": -15.0,
+ "whip_q": 16.9024,
+ "whip_ratio": 1.44994,
+ "pan_w": 0.103043,
+ "width": 0.0663294,
+ "room_wet": 0.18326,
+ "rt60": 0.159853,
+ "air_db": -12.0,
+ "floor_db": -72.8417,
+ "cut_db": -44.4128,
+ "cut_fade": 0.032451
+ },
+ "rank_score": 7.778,
+ "context": "a fast crisp air whoosh | whoosh_fast_01.wav,whoosh_fast_02.wav,whoosh_fast_03.wav,whoosh_fast_04.wav",
+ "total": 7.84,
+ "tone_db": 6.42,
+ "passes": true,
+ "source": "v1 forge-test final (demo/forge-test/whoosh/final)",
+ "robust_mean": 7.778,
+ "robust_pass_rate": 1.0
+ }
+ ]
+}
\ No newline at end of file
diff --git a/earshot/forge/tone.py b/earshot/forge/tone.py
index b045dda..2d0feb3 100644
--- a/earshot/forge/tone.py
+++ b/earshot/forge/tone.py
@@ -23,6 +23,37 @@ Measures (mono sum, the whole file):
range; length_s the file length
shares energy share of 5 broad regions (<150 Hz, 150 Hz-1 kHz, 1-4, 4-12, >12 kHz) in dB of the total, with the
refs' median share; region_penalty() counts the excess beyond median + 6 dB (a low puff, a bright hiss)
+
+Spectro-temporal envelope (EnvRef): the average spectrum above cannot hear a tail that dies too fast, a swell where
+the refs hit, or a band that rings on after the rest has gone (the blind test caught the forged impact by its tail).
+
+ from earshot.forge.tone import EnvRef
+ er = EnvRef(['real_01.wav', ...])
+ m = er.check(x)
+ m['env_db'], m['env_ratio'], m['env_out'], m['words'] # e.g. 'envelope 9.8 dB from the nearest ref
+ # kb_space_01.wav (the refs sit 6.3-8.1 apart); beyond every ref: tail 4-6.3 kHz +14 dB, tail 1.6-3.2 kHz +12 dB'
+ EnvRef.penalty(m)
+ bandgram 28 third-octave band energies (the CEN bands, 3rd-order Butterworth, so low bands smear in time like an
+ ear's) in 10 ms frames, dB re the loudest frame's total power, floored at -70 dB
+ env_dist the made sound's bandgram against one ref's: aligned at the onset (first frame within 20 dB of the
+ peak frame) or at the peak frame, whichever fits better, with a small tolerance (time-stretch 0.85-1.18
+ about the anchor, then a shift of +-30 ms around the best; stretch costs a little). Distance = mean
+ |dB difference| over every band x frame of both active spans (cells weighted by how loud the louder of
+ the two is, so a missing tail or an extra swell counts, silence does not; each cell's error capped at
+ 30 dB). The tolerance is deliberately small: a tail half as long as the ref's cannot be stretched away.
+ env_db distance to the NEAREST ref; env_ref: each ref's distance to its nearest other ref (what real spreads
+ to); env_ratio = env_db / median(env_ref). Nearest-ref distances flatter averages: with few, diverse refs
+ a sound in their middle sits closer to each than they sit to each other (the v1 pops, whooshes and
+ impacts do), so env_ratio alone cannot find a gap. Hence:
+ cells per segment (pre < -20 ms, attack -20..30 ms, body 30-200 ms, tail 0.2-1 s, late > 1 s from the anchor)
+ x 8 band groups (31-80 Hz ... 12.5-16 kHz): the signed weighted difference (made - ref) and the cell's
+ weight share, against each ref at its own best alignment
+ outside cells where the sound is beyond EVERY kept ref in the same direction by more than max(6 dB, the refs'
+ own median pairwise difference in that cell): a consistent gap that no real recording shares, in a cell
+ where the refs agree with each other. env_out = the sum of the excess dB beyond that tolerance, scaled
+ by min(1, share / 5 %); env_out_ref = the refs' own median env_out (each against the others; the cell
+ spreads use all refs, a small leak in leave-one-out scoring): diverse real refs are each unique
+ somewhere, so only the excess over it counts
"""
import numpy as np
@@ -170,3 +201,194 @@ class ToneRef:
if b > 0 and m['t40_ms'] > b:
pen += 0.5 * min(2.0, (m['t40_ms'] - b) / b)
return pen
+
+
+
+# ------------------------------------------------------------------------------- spectro-temporal envelope
+FLOOR = -70.0
+HOP_MS = 10.0
+STRETCH = (0.85, 0.92, 1.0, 1.08, 1.18)
+SHIFT = (-3, -2, -1, 0, 1, 2, 3) # frames (10 ms)
+SEGMENTS = (('pre', -1e9, -2), ('attack', -2, 3), ('body', 3, 20), ('tail', 20, 100), ('late', 100, 1e9)) # frames
+GROUPS = (('31-80 Hz', 0, 90), ('100-200 Hz', 90, 224), ('250-500 Hz', 224, 560), ('630-1250 Hz', 560, 1400),
+ ('1.6-3.2 kHz', 1400, 3550), ('4-6.3 kHz', 3550, 7100), ('8-10 kHz', 7100, 11200), ('12.5-16 kHz', 11200, 1e9))
+OUT_MARGIN = 6.0
+_SOS = []
+
+
+def _bank():
+ if not _SOS:
+ from scipy import signal
+ for c in CEN:
+ _SOS.append(signal.butter(3, [c / EDGE, min(c * EDGE, 0.49 * SR)], 'bandpass', fs=SR, output='sos'))
+ return _SOS
+
+
+def bandgram(x, hop_ms=HOP_MS):
+ """third-octave band energy over time: (frames, 28) dB re the loudest frame's total power, floored at FLOOR"""
+ from scipy import signal
+ m = _mono(x)
+ n = int(SR * hop_ms / 1000)
+ k = max(1, -(-len(m) // n))
+ m = np.pad(m, (0, k * n - len(m)))
+ G = np.empty((k, len(CEN)))
+ for j, sos in enumerate(_bank()):
+ G[:, j] = (signal.sosfilt(sos, m).reshape(k, n) ** 2).mean(1)
+ return np.maximum(10 * np.log10(G / (G.sum(1).max() + 1e-30) + 1e-30), FLOOR)
+
+
+def _total(G):
+ return 10 * np.log10((10 ** (G / 10)).sum(1) + 1e-30)
+
+
+def _anchors(G):
+ tot = _total(G)
+ ip = int(np.argmax(tot))
+ on = int(np.nonzero(tot >= tot[ip] - 20)[0][0])
+ act = np.nonzero(tot >= tot[ip] - 50)[0]
+ return {'onset': on, 'peak': ip}, int(act[0]), int(act[-1])
+
+
+def _rows(G, t):
+ """G sampled at fractional frames t (linear in dB), FLOOR outside"""
+ i0 = np.floor(t).astype(int)
+ fr = (t - i0)[:, None]
+ n = len(G)
+ a = np.where(((i0 >= 0) & (i0 < n))[:, None], G[np.clip(i0, 0, n - 1)], FLOOR)
+ b = np.where(((i0 + 1 >= 0) & (i0 + 1 < n))[:, None], G[np.clip(i0 + 1, 0, n - 1)], FLOOR)
+ return a * (1 - fr) + b * fr
+
+
+def env_dist(A, B, stretch=STRETCH, shift=SHIFT, detail=False):
+ """distance in dB between two bandgrams (A = the sound judged, B = a ref), best over the anchors / stretch /
+ shift tolerance; detail=True also returns the alignment and, per segment x band group, the signed weighted mean
+ difference (A - B, dB) and the cell's share of the weight: cells {'tail 630-1250 Hz': (diff, share)}"""
+ ancA, a_lo, a_hi = _anchors(A)
+ ancB, b_lo, b_hi = _anchors(B)
+ frames = {}
+ for name in ('onset', 'peak'):
+ a0, b0 = ancA[name], ancB[name]
+ lo = max(-300, min(a_lo - a0, b_lo - b0)) - 3
+ hi = min(600, max(a_hi - a0, b_hi - b0)) + 4
+ tt = np.arange(lo, hi, dtype=float)
+ frames[name] = (a0, tt, _rows(B, b0 + tt))
+
+ def cost(name, s, sh):
+ a0, tt, Bv = frames[name]
+ Av = _rows(A, a0 + tt / s + sh)
+ w = np.clip((np.maximum(Av, Bv) - FLOOR) / 40.0, 0.0, 1.0) # how loud the louder of the two is
+ d = float((w * np.minimum(np.abs(Av - Bv), 30.0)).sum() / (w.sum() + 1e-9))
+ return d * (1.0 + 0.5 * abs(np.log(s))), name, s, sh, tt, Av, Bv, w # stretching costs a little
+
+ best = min((cost(n, s, 0) for n in frames for s in stretch), key=lambda c: c[0]) # coarse: no shift
+ best = min([best] + [cost(best[1], best[2], sh) for sh in shift if sh], key=lambda c: c[0]) # then shift
+ d, name, s, sh, tt, Av, Bv, w = best
+ if not detail:
+ return d
+ cells, W = {}, w.sum() + 1e-9
+ for seg, t0, t1 in SEGMENTS:
+ rs = (tt >= t0) & (tt < t1)
+ for grp, f0, f1 in GROUPS:
+ cs = (CEN >= f0) & (CEN < f1)
+ ww = w[np.ix_(rs, cs)]
+ if ww.sum() < 0.01 * W:
+ continue
+ diff = float((ww * (Av[np.ix_(rs, cs)] - Bv[np.ix_(rs, cs)])).sum() / ww.sum())
+ cells[f'{seg} {grp}'] = (round(diff, 1), round(float(ww.sum() / W), 3))
+ return d, {'anchor': name, 'stretch': s, 'shift_ms': sh * HOP_MS, 'cells': cells}
+
+
+class EnvRef:
+ """the real references' band-energy trajectories, and a check of any sound's shape over time against them"""
+
+ def __init__(self, refs, names=None, margin=OUT_MARGIN):
+ xs = [_load(r) for r in refs]
+ self.names = list(names) if names is not None else [r if isinstance(r, str) else f'ref{i}' for i, r in
+ enumerate(refs)]
+ self.margin = margin
+ self.G = [bandgram(x) for x in xs]
+ self.n = len(self.G)
+ self.D = np.full((self.n, self.n), np.inf) # D[i, j] = env_dist(ref i judged against ref j)
+ det = {(i, j): env_dist(self.G[i], self.G[j], detail=True)
+ for i in range(self.n) for j in range(self.n) if i != j}
+ for (i, j), r in det.items():
+ self.D[i, j] = r[0]
+ pair = {} # how far the refs sit from each other, cell by cell
+ for r in det.values():
+ for key, (v, _) in r[1]['cells'].items():
+ pair.setdefault(key, []).append(abs(v))
+ self.cell_spread = {key: float(np.median(v)) for key, v in pair.items() if len(v) >= 2}
+ self.out_ref = np.array([self._outside([det[i, j] for j in range(self.n) if j != i])[1] for i in range(self.n)])
+
+ def _outside(self, res):
+ """cells beyond EVERY compared ref in one direction by more than the margin, and their weighted excess"""
+ outside, out_db = [], 0.0
+ spread = getattr(self, 'cell_spread', {})
+ if len(res) >= 2:
+ for key in res[0][1]['cells']:
+ got = [r[1]['cells'].get(key) for r in res]
+ if any(g is None for g in got):
+ continue
+ diffs = np.array([g[0] for g in got])
+ share = float(np.mean([g[1] for g in got]))
+ lo = diffs.min() if diffs.min() > 0 else (-diffs.max() if diffs.max() < 0 else 0.0)
+ tol = max(self.margin, spread.get(key, self.margin)) # where the refs disagree, allow as much
+ if lo > tol and share >= 0.02:
+ outside.append((key, round(float(np.sign(diffs[0]) * lo), 1), round(share, 3)))
+ out_db += (lo - tol) * min(1.0, share / 0.05)
+ return sorted(outside, key=lambda o: -abs(o[1]) * o[2]), float(out_db)
+
+ def spread(self, exclude=None):
+ """each kept ref's distance to its nearest OTHER kept ref (needs 3+ refs after the exclusion)"""
+ keep = [j for j in range(self.n) if j != exclude]
+ if len(keep) < 3:
+ return np.zeros(0)
+ return np.array([min(self.D[i, j] for j in keep if j != i) for i in keep])
+
+ def check(self, x, exclude=None, G=None):
+ """measure one sound (or its precomputed bandgram G) against the refs; exclude = a ref index to leave out"""
+ keep = [j for j in range(self.n) if j != exclude]
+ if not keep:
+ return {}
+ G = bandgram(_load(x)) if G is None else G
+ res = [env_dist(G, self.G[j], detail=True) for j in keep]
+ k = int(np.argmin([r[0] for r in res]))
+ d, info = res[k]
+ sp = self.spread(exclude)
+ ratio = d / max(float(np.median(sp)), 1.0) if len(sp) else None
+ outside, out_db = self._outside(res)
+ oref = self.out_ref[keep] if self.n >= 3 else np.zeros(0)
+ out = {'env_db': round(d, 2), 'env_ratio': None if ratio is None else round(ratio, 2),
+ 'env_ref': [round(float(v), 2) for v in sp],
+ 'nearest': self.names[keep[k]].replace(chr(92), '/').rsplit('/', 1)[-1], 'nearest_i': keep[k],
+ 'anchor': info['anchor'], 'stretch': info['stretch'], 'shift_ms': info['shift_ms'],
+ 'cells': info['cells'], 'outside': outside, 'env_out': round(out_db, 2),
+ 'env_out_ref': round(float(np.median(oref)), 2) if len(oref) else None}
+ out['words'] = self.words(out)
+ return out
+
+ @staticmethod
+ def words(m):
+ s = f"envelope {m['env_db']:.1f} dB from the nearest ref {m['nearest']}"
+ if m['env_ref']:
+ s += f" (the refs sit {min(m['env_ref']):.1f}-{max(m['env_ref']):.1f} apart)"
+ if m['outside']:
+ s += '; beyond every ref: ' + ', '.join(f'{k} {v:+.0f} dB' for k, v, _ in m['outside'][:3])
+ else:
+ worst = sorted(((k, v) for k, v in m['cells'].items() if abs(v[0]) >= 8),
+ key=lambda kv: -abs(kv[1][0]) * kv[1][1] ** 0.5)[:2]
+ if worst:
+ s += ': vs it ' + ', '.join(f'{k} {v[0]:+.0f} dB' for k, v in worst)
+ return s
+
+ @staticmethod
+ def penalty(m, cap=2.0):
+ """nearest part: 0 while the sound is no farther from its nearest ref than a typical ref from its nearest
+ neighbour, then linear in (env_ratio - 1); outside part: (env_out - max(3, the refs' own median env_out)) / 6,
+ so with diverse refs (each real one is unique somewhere) only an excess over what real recordings show
+ counts (6 dB beyond every ref in a cell holding 5 % or more of the weight = 1). Capped."""
+ if not m:
+ return 0.0
+ near = max(0.0, m['env_ratio'] - 1.0) if m.get('env_ratio') is not None else 0.0
+ tol = max(3.0, m.get('env_out_ref') or 0.0)
+ return float(min(cap, near + max(0.0, m.get('env_out', 0.0) - tol) / 6.0))
diff --git a/earshot/forge/viz.py b/earshot/forge/viz.py
new file mode 100644
index 0000000..b4e32ca
--- /dev/null
+++ b/earshot/forge/viz.py
@@ -0,0 +1,122 @@
+"""Eyes for the forge: made sounds next to their nearest real references, as one PNG.
+
+ from earshot.forge.viz import compare_png
+ compare_png('made/keypress/compare.png', [('keypress_01', x1), ('keypress_02', x2)],
+ [('kb_thock_02.wav', r2), ('kb_enter_01.wav', r5)], pairs=[0, 1]) # pairs: a ref index per made sound
+
+One row per made sound: its log-frequency spectrogram | its nearest reference's spectrogram (same time span, same
+80 dB range, both onset-aligned) | the two dB envelopes on one time axis (a tail that dies too fast shows here) |
+the 1/3-octave tonal balance of the made sound against every reference's (min-max band and median). `pairs=None`
+pairs each made sound with its nearest reference by earshot.forge.search.sound_dist (tonal balance + envelope).
+Returns {'file', 'rows', 'pairs'}. Needs matplotlib (imported lazily, Agg backend).
+"""
+import numpy as np
+
+SR = 48000
+
+
+def _mono(x):
+ x = np.asarray(x, np.float64)
+ return x.mean(axis=1) if x.ndim == 2 else x
+
+
+def onset_align(x, db=-30.0, pre_ms=5.0):
+ """the sound from `pre_ms` before its first sample within `db` of the peak"""
+ m = _mono(x)
+ on = int(np.argmax(np.abs(m) > np.abs(m).max() * 10 ** (db / 20)))
+ return m[max(0, on - int(SR * pre_ms / 1000)):]
+
+
+def spec_db(x, span_s):
+ """log-magnitude STFT, 0 dB = the loudest bin, floored at -80: (t_ms, f_hz, S_db)"""
+ from scipy import signal
+ m = np.pad(x, (0, max(0, int(span_s * SR) - len(x))))[:max(int(span_s * SR), 256)]
+ nfft = 512 if span_s < 0.6 else 1024 if span_s < 2.5 else 2048
+ f, t, S = signal.stft(m, SR, nperseg=nfft, noverlap=nfft - nfft // 8, boundary=None)
+ S = 20 * np.log10(np.abs(S) + 1e-12)
+ return t * 1000, f, np.maximum(S - S.max(), -80.0)
+
+
+def envelope_db(x, span_s, hop_ms=2.0):
+ """`hop_ms` RMS envelope in dB re its peak, floored at -80: (t_ms, e_db)"""
+ n = int(SR * hop_ms / 1000)
+ m = np.pad(x, (0, max(0, int(span_s * SR) - len(x))))[:int(span_s * SR)]
+ k = max(1, len(m) // n)
+ e = 10 * np.log10((m[:k * n].reshape(k, n) ** 2).mean(1) + 1e-18)
+ return np.arange(k) * hop_ms, np.maximum(e - e.max(), -80.0)
+
+
+def compare_png(path, made, refs=(), pairs=None, title=None, max_s=8.0):
+ """write the comparison sheet; made / refs = [(name, 48 kHz array), ...]; pairs = a ref index (or None) per made"""
+ import matplotlib
+ matplotlib.use('Agg')
+ import matplotlib.pyplot as plt
+ from .search import describe, sound_dist
+ from .tone import CEN, profile
+ made, refs = list(made), list(refs)
+ if pairs is None:
+ rd = [describe(x) for _, x in refs]
+ pairs = [int(np.argmin([sound_dist(describe(x), r) for r in rd])) if rd else None for _, x in made]
+ RP = np.stack([profile(x) for _, x in refs]) if refs else None
+ rows = max(1, len(made))
+ plt.rcParams.update({'font.size': 7.5, 'axes.titlesize': 8})
+ fig, axes = plt.subplots(rows, 4, figsize=(15, 2.25 * rows + (0.5 if title else 0.2)), squeeze=False,
+ gridspec_kw={'width_ratios': [1.25, 1.25, 1.0, 1.0]})
+ c_made, c_ref = '#1f77b4', '#d62728'
+ for i, (name, x) in enumerate(made):
+ a = onset_align(x)
+ j = pairs[i] if i < len(pairs) else None
+ b = onset_align(refs[j][1]) if j is not None else None
+ span = min(max_s, max(len(a), len(b) if b is not None else 0) / SR)
+ for col, (nm, y) in enumerate([(name, a), (refs[j][0] if j is not None else 'no reference', b)]):
+ ax = axes[i, col]
+ if y is None:
+ ax.axis('off')
+ continue
+ t, f, S = spec_db(y, span)
+ ax.pcolormesh(t, np.maximum(f, 20), S, shading='auto', cmap='magma', vmin=-80, vmax=0, rasterized=True)
+ ax.set_yscale('log')
+ ax.set_ylim(40, 24000)
+ ax.set_xlim(0, span * 1000)
+ ax.set_title(('made: ' if col == 0 else 'nearest ref: ') + nm, color=c_made if col == 0 else c_ref)
+ ax.set_ylabel('Hz')
+ if i == rows - 1:
+ ax.set_xlabel('ms from onset')
+ ax = axes[i, 2]
+ hop = max(2.0, span * 1000 / 300) # ~300 points per panel: long tails read as a line, not a band
+ t, e = envelope_db(a, span, hop)
+ ax.plot(t, e, color=c_made, lw=1.1, label='made')
+ if b is not None:
+ t2, e2 = envelope_db(b, span, hop)
+ ax.plot(t2, e2, color=c_ref, lw=1.1, label='ref')
+ ax.set_ylim(-70, 3)
+ ax.set_xlim(0, span * 1000)
+ ax.grid(alpha=0.25)
+ ax.set_title('envelope (dB re peak)')
+ ax.legend(loc='upper right', frameon=False)
+ ax = axes[i, 3]
+ if RP is not None:
+ ax.fill_between(CEN, RP.min(0), RP.max(0), color=c_ref, alpha=0.18, lw=0, label='refs min-max')
+ ax.plot(CEN, np.median(RP, 0), color=c_ref, lw=1.0, ls='--', label='refs median')
+ p = profile(x)
+ ax.plot(CEN, p, color=c_made, lw=1.3, label='made')
+ if RP is not None:
+ ax.set_title(f'1/3-octave balance: {np.sqrt(np.mean((p - np.median(RP, 0)) ** 2)):.1f} dB from the refs')
+ else:
+ ax.set_title('1/3-octave balance')
+ ax.set_xscale('log')
+ ax.set_xlim(25, 20000)
+ ax.set_ylim(-60, 0)
+ ax.grid(alpha=0.25)
+ if i == 0:
+ ax.legend(loc='lower left', frameon=False)
+ for i in range(len(made), rows):
+ for ax in axes[i]:
+ ax.axis('off')
+ if title:
+ fig.suptitle(title, fontsize=10)
+ fig.tight_layout()
+ fig.savefig(path, dpi=90)
+ plt.close(fig)
+ return {'file': str(path).replace(chr(92), '/'), 'rows': len(made),
+ 'pairs': [refs[j][0] if j is not None else None for j in pairs]}
diff --git a/eval/forge/RESULTS.md b/eval/forge/RESULTS.md
new file mode 100644
index 0000000..025c988
--- /dev/null
+++ b/eval/forge/RESULTS.md
@@ -0,0 +1,165 @@
+# Forge benchmark (eval/forge)
+
+A fixed yardstick for sounds the forge makes, so the overnight rounds cannot fool themselves. The forge optimises
+against **forge** refs; this benchmark measures against **bench** refs: real / pro recordings from source files (and
+authors) the forge never sees, using checks no recipe optimises.
+
+```
+python eval/forge/build_refsets.py # dev machine: cut the refs, write <EARSHOT_HOME>/refsets.json
+python eval/forge/bench.py --made made/ --out eval/forge/results_<tag>.json # made/<category>/*.wav
+python eval/forge/bench.py --files typing_key="demo/forge-test/keypress/final/*.wav" ... --out ...
+python eval/forge/bench.py --real-only --out eval/forge/results_real.json # only the real ceilings
+python eval/forge/bench.py --out eval/forge/results_<tag>.json --retable # rewrite the .md from the JSON
+```
+
+Each run writes a JSON (every number per sound) and a markdown table next to it. Run it with `EARSHOT_THREADS=2`. With
+the embeddings already cached, the v1 run took 12-25 s and the real-only run of all 18 categories 19-49 s. The first
+real-only run, with nothing cached, took 14 min.
+
+## Reference sets (`earshot.forge.refsets`)
+
+`refset(category, split='forge' | 'bench')` returns the paths from the local registry (`EARSHOT_REFSETS`, otherwise
+`<EARSHOT_HOME>/refsets.json`). If the registry is missing, it raises a clear error, because public users bring their
+own refs: put them in `<root>/<category>/{forge,bench}/*.wav`, then run `refsets.write(refsets.scan(root))`.
+
+The refs on this machine (`out/refs/<category>/{forge,bench}/`, with `out/refs/manifest.json`) are cut from
+`demo/launch-sfx-kit/sources.json` recordings. They come either from the kit's own trims (re-cut from the source,
+without the kit's high-pass) or from `earshot find` one-shot segments with clean edges and z >= 2. Processing is trim,
+fades and peak gain to -1 dBFS, nothing else. The manifest stores each trim's source, trim, licence and earshot
+evidence (find z, rank z of the category query, t40, length), plus every rejected candidate and the reason.
+
+**Split rule.** A source *family* is either forge or bench, never both, in any category. On Freesound the family is
+the author, which keeps sister takes and sample packs together. VSCO counts as one library; Mixkit goes file by file.
+The build refuses to write a registry that breaks this rule. Sources the v1 recipes used as refs are on the forge side.
+
+| category | forge refs (source files) | bench refs (source files) | rank z of the category query, median forge / bench | licences |
+|---|---|---|---|---|
+| whoosh_fast | 5 (4) | 5 (3) | 4.35 / 4.19 | CC0, CC BY |
+| whoosh_big | 3 (3) | 3 (2) | 4.67 / 3.58 | CC0, CC BY, Mixkit |
+| swish | 7 (3) | 3 (3) | 3.31 / 3.62 | CC0, CC BY |
+| swipe | 4 (1) | 4 (1) | 4.40 / 4.51 | CC0, CC BY |
+| riser | 1 (1) | 2 (2) | 2.66 / 3.42 | CC0, Mixkit |
+| downlifter | 0 | 0 | - | - |
+| impact | 2 (2) | 1 (1) | 3.67 / 3.26 | CC0, CC BY, Mixkit |
+| sub_drop | 0 | 1 (1) | - / 4.55 | CC0 |
+| click | 8 (5) | 7 (5) | 2.69 / 2.61 | CC0, CC BY, Mixkit |
+| tap | 7 (2) | 6 (2) | 1.69 / 1.62 | CC0 |
+| toggle | 4 (4) | 4 (3) | 2.10 / 1.92 | CC0, CC BY, Mixkit |
+| hover_tick | 6 (2) | 6 (3) | 1.73 / 1.48 | CC0 |
+| pop | 7 (6) | 6 (6) | 3.08 / 2.88 | CC0, CC BY, Mixkit |
+| notify | 5 (5) | 5 (5) | 1.27 / 1.93 | CC0, CC BY |
+| shimmer | 5 (4) | 4 (4) | 3.37 / 4.16 | CC0, CC BY, Mixkit |
+| typing_key | 15 (10) | 12 (9) | 5.13 / 3.94 | CC0, CC BY |
+| glitch | 0 | 0 | - | - |
+| logo_sting | 2 (2) | 2 (2) | 2.38 / 1.87 | CC0, Mixkit |
+
+Total: 81 forge refs from 40 families, 71 bench refs from 37 families.
+
+Thin or empty sets, honestly:
+- **downlifter** and **glitch**: no material in the library, real or designed. Made sounds in these categories get
+ only the phrasings, AudioSet and artifact checks.
+- **sub_drop**: one designed source, held out as bench. The forge has no sub-drop refs.
+- **impact**: the library has 3 cinematic hits, and v1 used all three as refs. One (`kit impact_03`) is now the only
+ bench ref, so the v1 impact numbers are optimistic and typicality is undefined (it needs 3 refs).
+- **riser**: one designed trailer riser (forge); the bench is two real cymbal swells.
+- **logo_sting**: stings are composed, so there is no real one. The refs are struck metal / glass rings, useful for
+ tail shape and tone only.
+- **swipe**: cloth only, no paper.
+- **Weak CLAP evidence**: tap, hover_tick and notify bench refs have rank z around 1.5-1.9. CLAP hears finger taps,
+ ticks and kalimba / celesta "dings" only weakly.
+
+**Kit WAVs that come from bench sources.** A forge recipe must not use these as refs: `ui_notify_02`,
+`whoosh_med_01`, `whoosh_med_02`, `whoosh_big_02`, `swipe_01`, `riser_02`, `impact_03`, `shimmer_01`. Use
+`refset(cat)` (forge split) instead of kit paths. In rounds 1-11, v1's whoosh used `whoosh_med_01/02`; one bench take
+(`fs-463763` 1.217-1.615 s) overlaps the `whoosh_med_01` trim, so v1's whoosh numbers are slightly optimistic too.
+
+## What the benchmark measures (all against the category's bench refs)
+
+| check | what | better |
+|---|---|---|
+| typ z | CLAP typicality: closeness to the nearest 3 bench refs, compared with how close the refs sit to each other (needs 3+ refs) | higher |
+| tone dB | 1/3-octave tonal-balance distance to the bench median (`frozen_tone.py`: `earshot/forge/tone.py` frozen at round 0, so the forge's evolving copy never moves the benchmark) | lower |
+| env oct | mean \|log2(made / bench median)\| of attack, t40 and active length (1 = typically 2x off) | lower |
+| curve dB | RMS dB distance of the peak-aligned 10 ms envelope to the bench median envelope: where a tail that dies too fast shows | lower |
+| t40 x | tail length / bench median | ~1 |
+| phrase z, margin, wins | CLAP z of 2 independent phrasings per category, never a recipe TARGET / ALT (checked at start); margin over 4 category confusions | higher |
+| AS names, as_p | an AudioSet top-3 label names the category; best label probability | higher |
+| artifacts | clipped, dc offset, start click, end click, tail cut, late start (calibrated: 5 of 150 real trims trip one, all ticks cut from a flurry or plain 5 s windows) | none |
+| var cos | mean pairwise CLAP cosine among a category's variants (real sets: 0.48-0.86, logo_sting's two forge refs 0.19) | lower = more varied |
+| texture | floor dB (recording floor), stereo corr, darken (does the decay darken) | context |
+
+Real ceilings, scored by the same functions:
+- **real x**: the forge refs (real, but never in the bench) scored against the bench refs. This is the yardstick to
+ compare made sounds with.
+- **real loo**: each bench ref scored against the others.
+- **real_rev** (JSON only): the bench refs scored against the forge refs.
+
+**inside** counts, per made sound, how many checks land inside the real rows' 80% limits, shown next to the same
+count for the real rows.
+
+## Baseline: v1 finals (`results_v1.json`, `results_v1.md`)
+
+Mapping: keypress → typing_key, click → click, pop → pop, whoosh → whoosh_fast, impact → impact. Medians:
+
+| category | set | n | typ z | tone dB | env oct | curve dB | t40 x | phrase z | margin | AS names | artifacts | var cos | inside |
+|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
+| click | made | 3 | 1.89 | 7.6 | 0.56 | 8.85 | 0.56 | 3.12 | 1.13 | 0 | 0 | 0.886 | 7 / 7 (real 6) |
+| | real x | 8 | 0.16 | 11.33 | 0.22 | 9.92 | 1.33 | 2.77 | 0.53 | 0.12 | 0 | 0.651 | |
+| | real loo | 7 | -0.1 | 9.85 | 0.46 | 13.19 | 0.86 | 2.67 | 0.27 | 0 | 0 | 0.576 | |
+| impact | made | 3 | | 6.7 | 1.54 | 24.86 | 0.5 | 3.68 | 1.81 | 1 | 0 | 0.931 | 5 / 6 (real 3) |
+| | real x | 2 | | 6.19 | 2.5 | 31.98 | 0.46 | 3.04 | 1.09 | 0.5 | 0 | 0.659 | |
+| pop | made | 3 | 1.33 | 8.59 | 0.68 | 6.29 | 1 | 3.09 | -0.1 | 1 | 0 | 0.917 | 7 / 7 (real 6) |
+| | real x | 7 | -0.09 | 16.06 | 1.51 | 20.54 | 0.91 | 2.46 | -0.5 | 0.29 | 0 | 0.501 | |
+| | real loo | 6 | 1.13 | 13.66 | 0.41 | 12.5 | 1.02 | 1.96 | -1.06 | 0.67 | 0 | 0.522 | |
+| typing_key | made | 3 | 1.2 | 8.2 | 1.07 | 14.32 | 1.67 | 5.56 | 3.57 | 0.33 | 0 | 0.965 | 7 / 7 (real 6) |
+| | real x | 15 | 0.26 | 8.88 | 1.01 | 12.2 | 1.29 | 4.75 | 2.83 | 0.53 | 0 | 0.721 | |
+| | real loo | 12 | 0.23 | 5.47 | 1.52 | 16.7 | 1 | 3.71 | 1.53 | 0.42 | 0 | 0.673 | |
+| whoosh_fast | made | 3 | 2.05 | 8.58 | 0.89 | 15.93 | 0.51 | 4.46 | 3.31 | 0.33 | 0 | 0.957 | 7 / 7 (real 6.5) |
+| | real x | 5 | 0.57 | 10.31 | 1.22 | 20.2 | 0.6 | 4.14 | 2.78 | 0.4 | 0 | 0.826 | |
+| | real loo | 5 | 1.52 | 4.58 | 0.4 | 8.21 | 1.08 | 4.14 | 2.62 | 0.8 | 0 | 0.772 | |
+
+How to read it:
+1. **Per sound, v1 cannot be told from real on these checks.** Each made sound lands inside the real 80% limits on
+ 7 of 7 checks (impact 5 of 6), while a typical real recording lands inside on 6. This includes the model-free tone
+ and envelope distances against held-out refs. The reason: real recordings from different sources sit far from each
+ other (real x tone 9-16 dB, curve 10-32 dB), so "as close as another real recording" is a loose bar. This is the
+ same saturation the forge-test judge found, now measured on a held-out set.
+2. **The clear made-vs-real gap is variety.** The 3 variants of each made category sit at pairwise CLAP cosine 0.89-0.97;
+ real sets sit at 0.50-0.83. The forge's "diverse" picks sound alike. A user who hears three made clicks in a row
+ will notice it.
+3. **Made sounds are prototypes.** In every category, the made median typ z is above both real medians: click 1.89
+ vs 0.16 / -0.1, pop 1.33 vs -0.09 / 1.13, typing_key 1.2 vs 0.26 / 0.23, whoosh_fast 2.05 vs 0.57 / 1.52. They sit
+ nearer the centre of the held-out cluster than real recordings from other sources do. This fits a search pulled toward
+ CLAP closeness. It is not a defect by itself, but it means typ z and phrase z cannot be pushed higher to show
+ progress: they are already above real.
+4. **Tails.** Made clicks have t40 0.56x the bench median, against the real refs' 0.86-1.33. Made whooshes are at
+ 0.51x, against 0.6-1.08. Both are short tails. Made impacts are at 0.35-0.62x the single held-out impact's 3.45 s
+ tail. But the two forge impacts are just as short (0.38x, 0.54x), so one bench ref cannot settle it. The user's ear
+ ("the tail died too fast") points the same way.
+5. **Phrasings are saturated.** Phrase z and margin are above real in every category, as expected, so don't read
+ progress from them. AudioSet naming is mixed: made sounds beat real for pop (1 vs 0.29 / 0.67) and impact (1 vs
+ 0.5), but fall below it for click (0 vs 0.12 / 0), typing_key (0.33 vs 0.53 / 0.42) and whoosh_fast (0.33 vs 0.4 /
+ 0.8). AudioSet rarely names a UI click at all.
+
+So for the overnight rounds, progress is shown by: lower var cos (towards 0.5-0.8) with no tone / env regressions;
+t40 x moving towards 1 (clicks, whooshes); tone / curve distances at or under real x; zero artifacts; and the new
+categories getting rows at all. The final word is the blind test.
+
+## Limits
+
+- The sets are small: 1-15 refs per split, and 1-3 source files in several categories. Real-limit quantiles from 2-3
+ rows are rough.
+- Real refs are trims cut by earshot find with fades. Their tails and floors partly reflect the cut. `floor dB` on
+ short trims is mostly the trim's own tail, not a room floor.
+- tone dB and the envelope checks come from the same family of measurements the forge can optimise (`tone=`). On
+ bench refs they are still held out, but they are not independent of the forge's own ears in kind.
+- AudioSet has no class for riser, downlifter or glitch; the label sets use the nearest classes. AudioSet hears only
+ up to 8 kHz.
+- Judge sounds by these numbers plus spectrograms plus ears. Never select or steer per candidate by the benchmark: it
+ is held out only while it decides nothing.
+
+## Benchmark changes (only additions, recorded)
+
+| when | change |
+|---|---|
+| 2026-10-16 round 0 | refsets v1 (18 categories, 152 refs); bench.py v1 (checks above); frozen_tone.py from git 108763c; baseline results_v1 + results_real |
diff --git a/eval/forge/bench.py b/eval/forge/bench.py
new file mode 100644
index 0000000..5f5b18e
--- /dev/null
+++ b/eval/forge/bench.py
@@ -0,0 +1,511 @@
+"""The fixed forge benchmark: made sounds per category against HELD-OUT real refs, with checks the forge never optimises.
+
+ python eval/forge/bench.py --made made/ --out eval/forge/results_round3.json # made/<category>/*.wav
+ python eval/forge/bench.py --files typing_key="demo/forge-test/keypress/final/*.wav" click=... --out ...
+ python eval/forge/bench.py --real-only --out eval/forge/results_real.json # only the real ceilings
+
+Refs: earshot.forge.refsets (the 'bench' split: source recordings the forge never sees; see eval/forge/build_refsets.py).
+Every made sound gets, against its category's bench refs (no model in the objective of any forge recipe here):
+ typ_z CLAP typicality: closeness to the nearest 3 bench refs vs how close the bench refs sit to each other
+ (compare._typicality; None with fewer than 3 refs)
+ tone_db 1/3-octave tonal balance distance to the bench refs' median (eval/forge/frozen_tone.py: earshot/forge/
+ tone.py frozen at round 0, so the forge's own evolving tone.py never moves the benchmark), tone_ratio = that /
+ the refs' own leave-one-out distance, n_outside = bands beyond every ref by > 3 dB
+ env_oct envelope shape: mean |log2(made / bench median)| of attack, t40 (peak to -40 dB) and active length
+ (1.0 = typically 2x off); curve_db = RMS dB distance of the peak-aligned 10 ms envelope (floored at -60 dB)
+ to the bench median envelope, which is where a tail that dies too fast shows
+ phrase_z calibrated CLAP z of 2-3 independent phrasings per category (PHRASES, never a recipe's TARGET / ALTS:
+ checked at start), margin = phrase z minus the best of the category's CONFUSIONS, wins = share of
+ phrasings with z >= 2.5 and above every confusion
+ as_names whether an AudioSet top-3 label names the category (AS_LABELS), as_p its best label probability
+ artifacts clipped, dc offset, start click (first sample non-zero), end click / tail cut (still sounding at the
+ end; not for risers, which stop on their peak), late start (nothing within 50 dB of the peak in the
+ first 100 ms: bad for frame sync); calibrated on the real trims (see artifacts())
+ diversity among a category's variants: CLAP cosine mean / max, tonal-balance distance min, near duplicates
+ (var cos in the table, next to the forge refs' and the bench refs' own: real sets sit ~0.5-0.8)
+Context ("real ceiling"), with the very same functions: real_cross = the category's FORGE refs scored against the
+bench refs (real recordings the bench never saw, the same yardstick as the made sounds); real_loo = each bench ref
+against the other bench refs; real_rev = the bench refs against the forge refs (stored in the JSON only).
+texture floor_db (the recording floor; digital silence = -120), stereo_corr, darken (does the decay darken)
+inside = per made sound, how many checks land inside the real rows' 80% limits (the 80th percentile of real_cross +
+real_loo for lower-is-better checks, the 20th for higher-is-better), next to the same count for the real rows.
+Writes the JSON (--out) and a markdown table next to it (.md). Models: CLAP + AudioSet (EARSHOT_THREADS=2 is gentle).
+"""
+import argparse
+import glob
+import json
+import os
+import sys
+import time
+
+import numpy as np
+
+HERE = os.path.dirname(os.path.abspath(__file__))
+REPO = os.path.dirname(os.path.dirname(HERE)).replace(chr(92), '/')
+for _p in (REPO, HERE):
+ if _p not in sys.path:
+ sys.path.insert(0, _p)
+import frozen_tone as tone # noqa: E402 (a frozen copy of earshot/forge/tone.py: the benchmark must not drift)
+
+SR = 48000
+PHRASES = { # independent of the recipes' TARGET / ALTS and of build_refsets' find queries
+ 'whoosh_fast': ['a quick swoosh as something flies past', 'a short sharp swoosh of air'],
+ 'whoosh_big': ['a heavy swooping rush of air', 'a big slow swoosh in a movie trailer'],
+ 'swish': ['a thin airy swish', 'a light quick swoosh of a thrown object'],
+ 'swipe': ['fabric rustling in one quick swipe', 'a sleeve brushing past quickly'],
+ 'riser': ['a tension build-up getting louder and louder', 'a whooshing crescendo that swells to a climax'],
+ 'downlifter': ['a descending whoosh that fades out', 'a falling pitch sweep winding down'],
+ 'impact': ['a huge low thud for a movie trailer', 'a massive hit with a rumbling tail'],
+ 'sub_drop': ['a low frequency boom that drops in pitch', 'a deep rumbling bass drop'],
+ 'click': ['a crisp tiny plastic click', 'a short clean click of a button being pressed'],
+ 'tap': ['a fingertip tapping on glass', 'a light tap of a finger on a phone screen'],
+ 'toggle': ['a light switch clicking on', 'a small switch being flipped'],
+ 'hover_tick': ['a very short faint tick', 'a tiny dry click'],
+ 'pop': ['a popping sound like a bubble bursting', 'a quick plop'],
+ 'notify': ['a pleasant chime alert sound', 'a single bell ding on a phone'],
+ 'shimmer': ['sparkling twinkling chimes', 'a glittering magical sound'],
+ 'typing_key': ['one key being pressed on a computer keyboard', 'a single keystroke on a mechanical keyboard'],
+ 'glitch': ['a digital glitch with stuttering noise', 'a broken electronic signal stutter'],
+ 'logo_sting': ['a short musical logo with a ringing finish', 'a bright struck bell ringing out'],
+}
+CONFUSIONS = {
+ 'whoosh_fast': ['a steam hiss', 'a car driving past', 'tearing paper', 'a bird flapping its wings'],
+ 'whoosh_big': ['a strong wind storm', 'an airplane taking off', 'ocean waves crashing', 'a car driving past'],
+ 'swish': ['a short hiss', 'a sword being drawn', 'a page being turned', 'a bird flying past'],
+ 'swipe': ['paper crumpling', 'a broom sweeping', 'a gust of wind', 'a zipper'],
+ 'riser': ['a jet engine starting', 'a siren', 'steady rain', 'an orchestra tuning'],
+ 'downlifter': ['a siren', 'a car passing by', 'a falling bomb whistle', 'wind'],
+ 'impact': ['a gunshot', 'a door being slammed', 'a bass drum', 'distant fireworks'],
+ 'sub_drop': ['an earthquake rumble', 'a kick drum', 'a truck engine idling', 'thunder in the distance'],
+ 'click': ['a wooden stick hit', 'a finger snap', 'a drop of water', 'a stapler'],
+ 'tap': ['a knock on a door', 'a raindrop', 'a keyboard key', 'a ballpoint pen click'],
+ 'toggle': ['a door latch', 'a stapler', 'a seatbelt buckle', 'a ball bouncing'],
+ 'hover_tick': ['a clock ticking', 'a drop of water', 'a typewriter', 'a cricket chirp'],
+ 'pop': ['a hand clap', 'a gunshot', 'a kiss', 'a tongue click'],
+ 'notify': ['a doorbell', 'a microwave beep', 'a church bell', 'a cash register'],
+ 'shimmer': ['a harp glissando', 'breaking glass', 'a music box', 'a sleigh bell jingle'],
+ 'typing_key': ['a ballpoint pen click', 'a stapler', 'a light switch', 'a dripping tap'],
+ 'glitch': ['radio static', 'a modem dialing', 'a scratched record', 'an electric buzz'],
+ 'logo_sting': ['a church bell', 'a doorbell', 'a gong', 'a cymbal crash'],
+}
+AS_LABELS = { # AudioSet classes that name the category (riser / downlifter / glitch have none exact: nearest)
+ 'whoosh_fast': ['Whoosh, swoosh, swish', 'Whip'], 'whoosh_big': ['Whoosh, swoosh, swish'],
+ 'swish': ['Whoosh, swoosh, swish', 'Whip'], 'swipe': ['Whoosh, swoosh, swish', 'Rustle'],
+ 'riser': ['Whoosh, swoosh, swish', 'Cymbal', 'Drum roll'], 'downlifter': ['Whoosh, swoosh, swish'],
+ 'impact': ['Boom', 'Thump, thud', 'Explosion', 'Bang', 'Bass drum'], 'sub_drop': ['Boom', 'Rumble', 'Thump, thud'],
+ 'click': ['Clicking', 'Tick', 'Tap'], 'tap': ['Tap', 'Knock', 'Clicking', 'Tick'],
+ 'toggle': ['Clicking', 'Mechanisms', 'Tick'], 'hover_tick': ['Tick', 'Clicking', 'Tick-tock'],
+ 'pop': ['Burst, pop', 'Plop'],
+ 'notify': ['Ding', 'Chime', 'Bell', 'Glockenspiel', 'Jingle, tinkle', 'Ding-dong', 'Ringtone', 'Marimba, xylophone',
+ 'Vibraphone'],
+ 'shimmer': ['Chime', 'Wind chime', 'Jingle, tinkle', 'Glockenspiel', 'Bell', 'Jingle bell'],
+ 'typing_key': ['Computer keyboard', 'Typing', 'Typewriter', 'Clicking'],
+ 'glitch': ['Static', 'Noise', 'Beep, bleep', 'White noise', 'Crackle'],
+ 'logo_sting': ['Chime', 'Bell', 'Ding', 'Jingle, tinkle', 'Gong', 'Cymbal', 'Glockenspiel'],
+}
+METRICS = (('typ_z', 'high'), ('tone_db', 'low'), ('env_oct', 'low'), ('curve_db', 'low'), ('phrase_z', 'high'),
+ ('margin', 'high'), ('as_p', 'high'))
+
+
+def _mono(x):
+ x = np.asarray(x, np.float64)
+ return x.mean(axis=1) if x.ndim == 2 else x
+
+
+def _frames_db(m, ms):
+ n = max(1, int(SR * ms / 1000))
+ k = max(1, len(m) // n)
+ return 20 * np.log10(np.sqrt((np.pad(m, (0, max(0, k * n - len(m))))[:k * n].reshape(k, n) ** 2).mean(1)) + 1e-12)
+
+
+# ------------------------------------------------------------------------------------------- model-free checks
+def env_feats(x):
+ """onset (first 1 ms frame within 20 dB of the peak), attack (onset to peak), t40 (tone.t40_ms), active length
+ (onset to the last 5 ms frame above peak - 40 dB), peak time; seconds / ms"""
+ m = _mono(x)
+ e1 = _frames_db(m, 1.0)
+ p1 = int(np.argmax(e1))
+ on = int(np.nonzero(e1 >= e1[p1] - 20)[0][0])
+ e5 = _frames_db(m, 5.0)
+ last = int(np.nonzero(e5 >= e5.max() - 40)[0][-1])
+ return {'onset_s': round(on / 1000, 4), 'attack_ms': float(max(1, p1 - on)), 't40_ms': tone.t40_ms(m),
+ 'active_s': round(max(0.005, (last + 1) * 0.005 - on / 1000), 3), 'peak_s': round(p1 / 1000, 4)}
+
+
+def _curve(x, pre, post):
+ e = _frames_db(_mono(x), 10.0)
+ p = int(np.argmax(e))
+ e = np.maximum(e - e[p], -60.0)
+ out = np.full(pre + post, -60.0)
+ a, b = max(0, p - pre), min(len(e), p + post)
+ out[pre - (p - a):pre + (b - p)] = e[a:b]
+ return out
+
+
+class EnvRef:
+ """envelope shape of a ref set and the distance of any sound to it (see the module docstring: env_oct, curve_db)"""
+ PRE = 10 # 100 ms before the peak
+
+ def __init__(self, refs):
+ xs = [_load(r) for r in refs]
+ self.n = len(xs)
+ self.F = [env_feats(x) for x in xs]
+ tail = max([f['onset_s'] + f['active_s'] - f['peak_s'] for f in self.F], default=0.1)
+ self.post = int(min(800, max(10, np.ceil(1.25 * tail / 0.01))))
+ self.C = np.stack([_curve(x, self.PRE, self.post) for x in xs]) if xs else None
+
+ def check(self, x, exclude=None):
+ keep = [j for j in range(self.n) if j != exclude]
+ if not keep:
+ return {}
+ f = env_feats(x)
+ med = {k: float(np.median([self.F[j][k] for j in keep])) for k in ('attack_ms', 't40_ms', 'active_s')}
+ floor = {'attack_ms': 1.0, 't40_ms': 5.0, 'active_s': 0.01}
+ logs = {k: float(np.log2(max(f[k], floor[k]) / max(med[k], floor[k]))) for k in med}
+ c = _curve(x, self.PRE, self.post)
+ return {'attack_ms': f['attack_ms'], 't40_ms': f['t40_ms'], 'active_s': f['active_s'], 'onset_s': f['onset_s'],
+ 'env_oct': round(float(np.mean(np.abs(list(logs.values())))), 3),
+ 'curve_db': round(float(np.sqrt(np.mean((c - np.median(self.C[keep], 0)) ** 2))), 2),
+ 't40_ratio': round(2 ** logs['t40_ms'], 3), 'attack_ratio': round(2 ** logs['attack_ms'], 3),
+ 'active_ratio': round(2 ** logs['active_s'], 3), 'ref_median': {k: round(v, 3) for k, v in med.items()}}
+
+
+ENDS_ON_PEAK = ('riser',) # a riser grows out of silence and stops on its peak: no late-start / tail-cut flags
+
+
+def artifacts(x, cat=None):
+ """artifact flags and the numbers behind them: clipped (>= 3 samples at full scale), dc offset (mean > -30 dB re
+ the peak), start click (first sample > 5 % of the peak), end click (last 1 ms RMS > 2 %), tail cut (last 10 ms
+ within 30 dB of the loudest 5 ms), late start (lead_s: the first 1 ms frame within 50 dB of the peak comes after
+ 100 ms, so the sound cannot be placed on a frame). Calibrated so the real bench / forge trims rarely trip them."""
+ xa = np.asarray(x, np.float64)
+ m = _mono(xa)
+ pk = float(np.abs(m).max()) + 1e-12
+ e5 = _frames_db(m, 5.0)
+ e1 = _frames_db(m, 1.0)
+ out = {'clipped_samples': int((np.abs(xa) >= 0.999).sum()), 'dc_db': round(float(20 * np.log10(abs(m.mean()) / pk + 1e-12)), 1),
+ 'first_sample': round(abs(float(m[0])) / pk, 3), 'tail_db': round(float(_frames_db(m[-int(0.01 * SR):], 10.0)[0] - e5.max()), 1),
+ 'last_ms_rel': round(float(np.sqrt(np.mean(m[-48:] ** 2))) / pk, 3),
+ 'lead_s': round(int(np.nonzero(e1 >= e1.max() - 50)[0][0]) / 1000, 3),
+ 'peak_dbfs': round(float(20 * np.log10(float(np.abs(xa).max()) + 1e-12)), 2)}
+ fl = []
+ if out['clipped_samples'] >= 3:
+ fl.append('clipped')
+ if out['dc_db'] > -30:
+ fl.append('dc offset')
+ if out['first_sample'] > 0.05:
+ fl.append('start click')
+ if out['last_ms_rel'] > 0.02 and cat not in ENDS_ON_PEAK:
+ fl.append('end click')
+ if out['tail_db'] > -30 and cat not in ENDS_ON_PEAK:
+ fl.append('tail cut')
+ if out['lead_s'] > 0.1 and cat not in ENDS_ON_PEAK:
+ fl.append('late start')
+ out['flags'] = fl
+ return out
+
+
+def diversity(xs, E):
+ """variety among one category's variants: CLAP cosine, tonal-balance distance, near duplicates (cos > 0.985 and
+ tone < 1.5 dB apart)"""
+ n = len(xs)
+ if n < 2:
+ return {'n': n}
+ E = np.asarray(E, np.float64)
+ E = E / (np.linalg.norm(E, axis=1, keepdims=True) + 1e-12)
+ P = [tone.profile(x) for x in xs]
+ pairs = [(i, j) for i in range(n) for j in range(i + 1, n)]
+ cos = [float(E[i] @ E[j]) for i, j in pairs]
+ td = [float(np.sqrt(np.mean((P[i] - P[j]) ** 2))) for i, j in pairs]
+ return {'n': n, 'cos_mean': round(float(np.mean(cos)), 3), 'cos_max': round(max(cos), 3),
+ 'tone_min_db': round(min(td), 2), 'tone_mean_db': round(float(np.mean(td)), 2),
+ 'near_duplicates': [[i, j] for (i, j), c, t in zip(pairs, cos, td) if c > 0.985 and t < 1.5]}
+
+
+def texture(x):
+ """model-free texture: floor_db (10th percentile of the 5 ms frames re the loudest: a recording's room / mic floor,
+ -120 = digital silence), stereo_corr (L / R correlation, 1 = mono), darken (log2 of the spectral centroid 20 ms at
+ half the t40 tail over the centroid at the peak: real decays usually darken, < 0)"""
+ xa = np.asarray(x, np.float64)
+ m = _mono(xa)
+ e5 = _frames_db(m, 5.0)
+ out = {'floor_db': round(float(max(np.percentile(e5, 10) - e5.max(), -120.0)), 1)}
+ if xa.ndim == 2 and xa.shape[1] == 2:
+ a, b = xa[:, 0] - xa[:, 0].mean(), xa[:, 1] - xa[:, 1].mean()
+ den = np.sqrt((a * a).sum() * (b * b).sum())
+ out['stereo_corr'] = round(float((a * b).sum() / den), 3) if den > 0 else 1.0
+ else:
+ out['stereo_corr'] = 1.0
+ t40 = tone.t40_ms(m)
+ n = int(0.02 * SR)
+ p = int(np.argmax(np.abs(m)))
+ late = p + int(t40 / 2000 * SR)
+ if t40 >= 40 and late + n <= len(m):
+ f = np.fft.rfftfreq(n, 1 / SR)
+
+ def cen(seg):
+ F = np.abs(np.fft.rfft(seg * np.hanning(len(seg))))
+ return float((F * f).sum() / (F.sum() + 1e-12))
+ out['darken'] = round(float(np.log2(max(cen(m[late:late + n]), 20.0) / max(cen(m[p:p + n]), 20.0))), 2)
+ return out
+
+
+def inside_real(made, real, metrics=METRICS, q=0.8):
+ """per made row, how many metrics land inside the real rows' q-quantile limit (q = 0.8: the 80th percentile of a
+ lower-is-better metric, the 20th of a higher-is-better one; q = 1: the worst real value)"""
+ lim = {}
+ for k, d in metrics:
+ v = [r[k] for r in real if r.get(k) is not None]
+ if v:
+ lim[k] = (float(np.quantile(v, q if d == 'low' else 1 - q)), d)
+ per = [sum(1 for k, (b, d) in lim.items() if r.get(k) is not None and (r[k] <= b if d == 'low' else r[k] >= b))
+ for r in made]
+ return {'per_sound': per, 'n_checks': len(lim), 'limits': {k: round(float(b), 3) for k, (b, _) in lim.items()}}
+
+
+def _load(x):
+ if isinstance(x, str):
+ from earshot import audio
+ return audio.load(x)
+ return x
+
+
+# ------------------------------------------------------------------------------------------- the benchmark
+def recipe_phrases():
+ """every TARGET / ALT string of the forge recipes (the benchmark's phrasings must not be among them)"""
+ from earshot.forge import load_recipe, recipes
+ out = set()
+ for r in recipes():
+ try:
+ m = load_recipe(r)
+ except Exception: # a recipe being edited by someone else must not stop the benchmark
+ continue
+ out |= {s.strip().lower() for s in [getattr(m, 'TARGET', '')] + list(getattr(m, 'ALTS', [])) if s}
+ return out
+
+
+class Yardstick:
+ """one ref set's CLAP embeddings, tonal balance and envelope shape"""
+
+ def __init__(self, paths, xs, E):
+ self.paths, self.E = paths, E
+ self.tone = tone.ToneRef(xs) if xs else None
+ self.env = EnvRef(xs) if xs else None
+
+ def check(self, x, e, exclude=None):
+ from earshot import compare
+ keep = [j for j in range(len(self.paths)) if j != exclude]
+ r = {}
+ if len(keep) >= 3:
+ _, z, _ = compare._typicality(e, self.E[keep], 3)
+ r['typ_z'] = round(float(z), 2)
+ if keep:
+ r['nn_cos'] = round(float((self.E[keep] @ e).max()), 3)
+ tm = self.tone.check(x, exclude=exclude)
+ r.update(tone_db=tm['tone_db'], tone_ratio=tm['tone_ratio'], n_outside=len(tm['outside']),
+ outside=sorted(tm['outside'], key=lambda o: -abs(o[1]))[:3])
+ r.update(self.env.check(x, exclude=exclude))
+ return r
+
+
+def bench_category(cat, made_paths, reg, models, real=True, log=print):
+ from earshot import audio
+ v, t = models
+ e_ = reg['categories'].get(cat, {'forge': [], 'bench': [], 'note': ''})
+ B, F = list(e_.get('bench', [])), list(e_.get('forge', []))
+ allp = list(made_paths) + B + F
+ if not allp:
+ return {'n_bench': 0, 'n_forge': 0, 'note': e_.get('note', ''), 'made': [], 'summary': {}}
+ X = [audio.load(p) for p in allp]
+ from earshot import vocab
+ E = vocab.embed_files(allp)
+ phr, conf = PHRASES.get(cat, []), CONFUSIONS.get(cat, [])
+ Z = v.calibrate(E, v.embed_text(phr + conf))['z_q'] if phr else None
+ labs = [lab for lab in AS_LABELS.get(cat, []) if lab in t.index]
+ nm, nb = len(made_paths), len(B)
+ ys_b = Yardstick(B, X[nm:nm + nb], E[nm:nm + nb])
+ ys_f = Yardstick(F, X[nm + nb:], E[nm + nb:])
+
+ def row(i, ys, exclude=None):
+ x, e = X[i], E[i]
+ r = {'file': allp[i].replace(chr(92), '/'), 'length_s': round(len(x) / SR, 3)}
+ r.update(ys.check(x, e, exclude) if ys.paths else {})
+ if Z is not None:
+ zp, zc = Z[i, :len(phr)], Z[i, len(phr):]
+ best_c = float(zc.max()) if len(zc) else -9.0
+ r.update(phrase_z=round(float(zp.mean()), 2), phrase_min=round(float(zp.min()), 2),
+ margin=round(float(np.mean(zp - best_c)), 2),
+ wins=round(float(np.mean((zp >= 2.5) & (zp > best_c))), 2),
+ best_confusion=conf[int(np.argmax(zc))] if len(zc) else None)
+ p = t.probs_file(allp[i])
+ top = t.top(p, 3)
+ r.update(as_top=top, as_names=any(lab in labs for lab, _ in top),
+ as_p=round(max([t.prob_of(p, lab) for lab in labs], default=0.0), 3))
+ r.update(texture(x))
+ a = artifacts(x, cat)
+ r['artifacts'] = a['flags']
+ r['artifact_detail'] = {k: a[k] for k in ('clipped_samples', 'dc_db', 'first_sample', 'tail_db', 'lead_s',
+ 'peak_dbfs')}
+ return r
+
+ out = {'n_bench': nb, 'n_forge': len(F), 'note': e_.get('note', ''), 'bench_refs': B,
+ 'made': [row(i, ys_b) for i in range(nm)]}
+ out['diversity'] = diversity(X[:nm], E[:nm]) if nm else {'n': 0}
+ if real:
+ out['real_cross'] = [row(nm + nb + j, ys_b) for j in range(len(F))] if nb else []
+ out['real_loo'] = [row(nm + j, ys_b, exclude=j) for j in range(nb)] if nb >= 2 else []
+ out['real_rev'] = [row(nm + j, ys_f) for j in range(nb)] if F else []
+ out['diversity_bench'] = diversity(X[nm:nm + nb], E[nm:nm + nb])
+ out['diversity_forge'] = diversity(X[nm + nb:], E[nm + nb:])
+ realrows = out['real_cross'] + out['real_loo']
+ if nm and realrows:
+ out['inside'] = inside_real(out['made'], realrows)
+ out['inside']['real_per_sound'] = inside_real(realrows, realrows)['per_sound']
+ out['summary'] = {k: _summ(out.get(k, [])) for k in ('made', 'real_cross', 'real_loo')}
+ for k, d in (('made', 'diversity'), ('real_cross', 'diversity_forge'), ('real_loo', 'diversity_bench')):
+ if out['summary'].get(k) and (out.get(d) or {}).get('n', 0) >= 2:
+ out['summary'][k]['var_cos'] = out[d]['cos_mean']
+ log(f" {cat}: {nm} made, {nb} bench, {len(F)} forge refs" + (
+ f"; inside the real 80% limits on {np.median(out['inside']['per_sound']):g} of {out['inside']['n_checks']} "
+ f"checks (a real recording: {np.median(out['inside']['real_per_sound']):g})" if out.get('inside') else ''))
+ return out
+
+
+def _summ(rows):
+ if not rows:
+ return {}
+ s = {'n': len(rows)}
+ for k, _ in METRICS + (('tone_ratio', 'low'), ('t40_ratio', ''), ('wins', 'high'), ('floor_db', ''),
+ ('stereo_corr', ''), ('darken', '')):
+ v = [r[k] for r in rows if r.get(k) is not None]
+ if v:
+ s[k] = round(float(np.median(v)), 2)
+ s['as_names'] = round(float(np.mean([bool(r.get('as_names')) for r in rows])), 2)
+ s['artifacts'] = sum(len(r.get('artifacts', [])) for r in rows)
+ return s
+
+
+def gather(made=None, files=None):
+ """{category: [paths]} from a folder of <category>/*.wav and / or cat=glob specs"""
+ out = {}
+ if made:
+ for c in sorted(os.listdir(made)):
+ d = os.path.join(made, c)
+ if os.path.isdir(d):
+ ps = sorted(glob.glob(os.path.join(d, '*.wav')))
+ if ps:
+ out[c] = [p.replace(chr(92), '/') for p in ps]
+ for spec in files or []:
+ c, g = spec.split('=', 1)
+ out.setdefault(c, []).extend(sorted(p.replace(chr(92), '/') for p in glob.glob(g)))
+ return out
+
+
+def table(res):
+ """markdown: per category, the medians of made vs the real ceilings"""
+ cols = [('typ_z', 'typ z'), ('tone_db', 'tone dB'), ('env_oct', 'env oct'), ('curve_db', 'curve dB'),
+ ('t40_ratio', 't40 x'), ('phrase_z', 'phrase z'), ('margin', 'margin'), ('as_names', 'AS names'),
+ ('artifacts', 'artifacts'), ('var_cos', 'var cos')]
+ L = [f"# forge benchmark: {res['tag']}", '', f"made: {res.get('made_from') or '-'}; refs: {res['registry']}; "
+ f"{res['created']}", '',
+ 'Medians per set. made = the made sounds vs the bench refs; real x = the forge refs (real, never in the bench) '
+ 'vs the bench refs; real loo = each bench ref vs the other bench refs. typ z, phrase z, margin, AS names: '
+ 'higher is better; tone dB, env oct (octaves off in attack / t40 / length), curve dB: lower is better; t40 x '
+ '= tail length / bench median (1 = same). inside = checks (of n) a made sound lands inside the real '
+ '80% limit, median over the made sounds (real = the same count for the real rows). var cos = mean pairwise CLAP '
+ 'cosine within the set (made variants / forge refs / bench refs): high = the variants are near clones.', '',
+ '| category | set | n | ' + ' | '.join(c for _, c in cols) + ' | inside |',
+ '|---|---|---|' + '---|' * (len(cols) + 1)]
+ for cat, r in res['categories'].items():
+ ins = r.get('inside')
+ first = True
+ for k, name in (('made', 'made'), ('real_cross', 'real x'), ('real_loo', 'real loo')):
+ s = r.get('summary', {}).get(k) or {}
+ if not s:
+ continue
+ cells = ['' if s.get(c) is None else (f'{s[c]:g}' if isinstance(s[c], (int, float)) else str(s[c]))
+ for c, _ in cols]
+ inside = (f"{np.median(ins['per_sound']):g} / {ins['n_checks']} (real {np.median(ins['real_per_sound']):g})"
+ if k == 'made' and ins else '')
+ L.append(f"| {cat if first else ''} | {name} | {s['n']} | " + ' | '.join(cells) + f' | {inside} |')
+ first = False
+ if first:
+ L.append(f"| {cat} | no rows | 0 | " + ' | ' * len(cols) + ' |')
+ if r.get('note'):
+ L.append(f"| | note: {r['note']} | | " + ' | ' * len(cols) + ' |')
+ L += ['', 'Texture (medians; model-free, not in any forge objective): floor dB = 10th percentile 5 ms frame re the '
+ 'loudest (-120 = digital silence), stereo corr (1 = mono), darken = log2 centroid at half the tail / at the '
+ 'peak (< 0: the decay darkens).', '', '| category | set | floor dB | stereo corr | darken |', '|---|---|---|---|---|']
+ for cat, r in res['categories'].items():
+ for k, name in (('made', 'made'), ('real_cross', 'real x'), ('real_loo', 'real loo')):
+ s = r.get('summary', {}).get(k) or {}
+ if s:
+ L.append(f"| {cat if k == 'made' else ''} | {name} | {s.get('floor_db', '')} | {s.get('stereo_corr', '')} | "
+ f"{s.get('darken', '')} |")
+ L += ['', 'Diversity among the made variants (CLAP cosine mean / max, tonal distance min dB, near duplicates):', '']
+ for cat, r in res['categories'].items():
+ d = r.get('diversity') or {}
+ if d.get('n', 0) >= 2:
+ db = r.get('diversity_bench') or {}
+ L.append(f"- {cat}: cos {d['cos_mean']} / {d['cos_max']}, tone min {d['tone_min_db']} dB, "
+ f"{len(d['near_duplicates'])} near duplicate pair(s)"
+ + (f" (bench refs: cos {db['cos_mean']} / {db['cos_max']}, tone min {db['tone_min_db']} dB)"
+ if db.get('n', 0) >= 2 else ''))
+ return '\n'.join(L) + '\n'
+
+
+def main(argv=None):
+ ap = argparse.ArgumentParser(description=__doc__.split('\n')[0])
+ ap.add_argument('--made', help='folder with <category>/*.wav')
+ ap.add_argument('--files', nargs='*', help='category=glob (quote the glob)')
+ ap.add_argument('--out', required=True, help='results JSON (a .md table is written next to it)')
+ ap.add_argument('--tag', help='name in the report (default: from --out)')
+ ap.add_argument('--cats', nargs='*', help='only these categories')
+ ap.add_argument('--real-only', action='store_true', help='only the real ceilings of every category')
+ ap.add_argument('--no-real', action='store_true', help='skip the real ceilings (faster)')
+ ap.add_argument('--registry', help='refsets registry (default earshot.forge.refsets.registry_path())')
+ ap.add_argument('--retable', action='store_true', help='only rewrite the .md table from an existing --out JSON')
+ a = ap.parse_args(argv)
+ if a.retable:
+ with open(a.out, encoding='utf-8') as fh:
+ res = json.load(fh)
+ for r in res['categories'].values():
+ r.setdefault('summary', {})
+ md = os.path.splitext(a.out)[0] + '.md'
+ with open(md, 'w', encoding='utf-8') as fh:
+ fh.write(table(res))
+ return print(f'wrote {md}')
+ from earshot.forge import refsets
+ reg = refsets.load(a.registry)
+ clash = {c: [p for p in ps if p.strip().lower() in recipe_phrases()] for c, ps in PHRASES.items()}
+ clash = {c: v for c, v in clash.items() if v}
+ if clash:
+ raise SystemExit(f'benchmark phrasings coincide with recipe TARGET / ALTS (no longer independent): {clash}')
+ made = {} if a.real_only else gather(a.made, a.files)
+ cats = a.cats or (sorted(reg['categories']) if a.real_only else sorted(made))
+ unknown = [c for c in made if c not in PHRASES]
+ if unknown:
+ raise SystemExit(f'unknown categories {unknown}; known: {sorted(PHRASES)}')
+ from earshot import vocab
+ from earshot.models import ast
+ models = (vocab.get(), ast.get())
+ t0 = time.time()
+ res = {'tag': a.tag or os.path.splitext(os.path.basename(a.out))[0].replace('results_', ''),
+ 'made_from': a.made or a.files, 'registry': a.registry or refsets.registry_path(),
+ 'created': time.strftime('%Y-%m-%d %H:%M'), 'phrases': {c: PHRASES[c] for c in cats},
+ 'confusions': {c: CONFUSIONS[c] for c in cats}, 'as_labels': {c: AS_LABELS[c] for c in cats},
+ 'categories': {}}
+ for c in cats:
+ res['categories'][c] = bench_category(c, made.get(c, []), reg, models, real=not a.no_real)
+ res['seconds'] = round(time.time() - t0, 1)
+ os.makedirs(os.path.dirname(os.path.abspath(a.out)), exist_ok=True)
+ with open(a.out, 'w', encoding='utf-8') as fh:
+ json.dump(res, fh, indent=1, default=lambda o: o.tolist() if hasattr(o, 'tolist') else str(o))
+ md = os.path.splitext(a.out)[0] + '.md'
+ with open(md, 'w', encoding='utf-8') as fh:
+ fh.write(table(res))
+ print(f'wrote {a.out} and {md} ({res["seconds"]} s)')
+
+
+if __name__ == '__main__':
+ main()
diff --git a/eval/forge/build_refsets.py b/eval/forge/build_refsets.py
new file mode 100644
index 0000000..19d32ee
--- /dev/null
+++ b/eval/forge/build_refsets.py
@@ -0,0 +1,374 @@
+"""Build the forge's per-category reference sets from the licensed library (dev machine only; nothing is committed).
+
+ python eval/forge/build_refsets.py # every category (re-uses cached earshot finds)
+ python eval/forge/build_refsets.py --cats click tap # only these (merged into the existing manifest)
+ python eval/forge/build_refsets.py --plan # print the plan (sources per split), no audio work
+
+Sources: demo/launch-sfx-kit/sources.json (licences, paths into E:/clawd animation/audio/assets/) and the kit WAVs'
+own trims (kit.json: the source file + trim of every kit sound; re-cut here from the source with trim + fade only, no
+high-pass). Each source entry is cut one of four ways:
+ kit the kit's trim of the source (its fades)
+ find earshot find(path, query) one-shot segments: clean edges (no head / tail cut, except at the file's own
+ start / end), length inside the category's range, z >= 2 (weak or better; toggle 1.5: CLAP hears switches
+ only weakly), peak within 18 dB of the file's loudest candidate; best k by score
+ whole the one sound in the file: from 5 ms before the first frame within 30 dB of the peak until the last frame
+ above peak - 45 dB (+20 ms), capped at the category's max length (a capped tail gets a 0.25 s fade)
+ rise a crescendo: back from the loudness peak while within 35 dB of it (at most max length), until peak + 0.3 s
+Every cut gets fades (2 ms in; 30 ms out, 0.25 s when capped) and peak gain to -1 dBFS, nothing else. Evidence per
+WAV: earshot rank z of the category query (one CLAP read of the exported file), the find z / edges when found, the
+kit's own earshot evidence, the measured t40 tail and length.
+
+The split: every category lists its 'forge' and 'bench' sources explicitly. A source FAMILY (Freesound: the author,
+which keeps sister takes and sample packs together; VSCO: the library; Mixkit: the file) may be forge or bench, never both, in
+any category; the build refuses to write a registry that breaks this. Sources the v1 forge recipes used as refs (the
+kit WAVs named in earshot/forge/recipes/*.py) are on the forge side, so the bench is held out from v1 too, except
+where a category has nothing else (noted).
+
+Outputs: E:/earshot/out/refs/<category>/{forge,bench}/*.wav, out/refs/manifest.json (source, trim, licence, earshot
+evidence, rejected candidates and why) and the registry <EARSHOT_HOME>/refsets.json (earshot.forge.refsets).
+"""
+import argparse
+import json
+import os
+import sys
+
+import numpy as np
+
+HERE = os.path.dirname(os.path.abspath(__file__))
+REPO = os.path.dirname(os.path.dirname(HERE)).replace(chr(92), '/')
+sys.path.insert(0, REPO)
+from earshot import audio # noqa: E402
+from earshot.forge import refsets # noqa: E402
+sys.path.insert(0, HERE)
+from frozen_tone import t40_ms # noqa: E402
+
+KIT = REPO + '/demo/launch-sfx-kit'
+OUT = REPO + '/out/refs'
+SR = 48000
+CACHE = OUT + '/_work/find_cache.json'
+
+# category: query (earshot find / rank, used only to locate and evidence the trims; the benchmark's phrasings are
+# different), dur (min, max s), forge / bench entries: 'kit:<name>' | (source id, k, mode[, {start, end}]), note
+CATS = {
+ 'whoosh_fast': dict(query='a fast crisp whoosh of a stick swung through the air', dur=(0.2, 0.9),
+ forge=['kit:whoosh_fast_01', 'kit:whoosh_fast_02', 'kit:whoosh_fast_03', 'kit:whoosh_fast_04',
+ ('fs-719636', 1, 'whole')],
+ bench=[('fs-352719', 1, 'whole'), ('fs-216675', 1, 'whole'), ('fs-463763', 3, 'find')]),
+ 'whoosh_big': dict(query='a big deep heavy whoosh passing by', dur=(0.7, 3.2),
+ forge=['kit:whoosh_big_01', ('mk-1167', 1, 'whole'), ('mk-1168', 1, 'whole', {'start': 0.9})],
+ bench=['kit:whoosh_big_02', ('fs-711122', 3, 'find', {'start': 13.0, 'end': 19.5})],
+ note='forge = three designed Mixkit whooshes; bench = real air / cape swoops.'),
+ 'swish': dict(query='a quick light swish of something thrown', dur=(0.12, 0.55),
+ forge=[('fs-427979', 4, 'find'), ('fs-719637', 1, 'whole'), ('fs-425856', 2, 'find')],
+ bench=[('fs-463763', 4, 'find'), ('fs-216675', 1, 'whole'), ('fs-352719', 1, 'whole')]),
+ 'swipe': dict(query='a soft cloth swipe, fabric swished quickly', dur=(0.25, 1.0),
+ forge=['kit:swipe_02', 'kit:whoosh_med_03', ('fs-425856', 2, 'find')],
+ bench=['kit:swipe_01', ('fs-495390', 3, 'find')],
+ note='cloth only: the library has no paper slides.'),
+ 'riser': dict(query='a rising swell building up to a loud peak', dur=(1.5, 6.0),
+ forge=['kit:riser_01'],
+ bench=['kit:riser_02', ('fs-425434', 1, 'rise')],
+ note='thin: one designed trailer riser (forge); the bench is two real cymbal swells, no synth risers.'),
+ 'downlifter': dict(query='a falling downlifter sweep sinking away', dur=(1.0, 4.0), forge=[], bench=[],
+ note='empty: the licensed library has no downlifter, real or designed (no reversing or pitching '
+ 'of other sounds here: trim + fade only). Benchmark made downlifters on phrasings, '
+ 'AudioSet and artifacts only.'),
+ 'impact': dict(query='a deep cinematic boom impact hit', dur=(1.0, 8.0),
+ forge=['kit:impact_01', 'kit:impact_02'],
+ bench=['kit:impact_03'],
+ note='thin: the library has three cinematic hits (all three were the v1 impact forge refs); one '
+ 'is held out as bench, so v1 impact bench numbers are optimistic.'),
+ 'sub_drop': dict(query='a deep sub bass drop', dur=(1.0, 4.0),
+ forge=[], bench=[('fs-428073', 1, 'whole')],
+ note='one designed source (a tom pitched down 24 semitones by its author), held out as bench; '
+ 'the forge has no sub-drop refs.'),
+ 'click': dict(query='a small plastic button click', dur=(0.03, 0.35),
+ forge=['kit:ui_click_01', 'kit:ui_click_02', 'kit:ui_click_03', 'kit:ui_click_04',
+ ('fs-167829', 2, 'find'), ('fs-383863', 2, 'find')],
+ bench=[('mk-1124', 1, 'whole'), ('mk-2364', 1, 'whole'), ('fs-610524', 2, 'find'),
+ ('fs-461799', 2, 'find'), ('fs-431419', 2, 'find')]),
+ 'tap': dict(query='a soft finger tap on a hard surface', dur=(0.03, 0.3),
+ forge=['kit:ui_click_01', ('fs-704862', 3, 'find'), ('fs-452621', 3, 'find')],
+ bench=[('fs-631822', 3, 'find'), ('fs-707827', 3, 'find')]),
+ 'toggle': dict(query='a toggle switch flipped with a crisp click', dur=(0.05, 0.45), zmin=1.5,
+ forge=['kit:ui_toggle_01', 'kit:ui_toggle_02', ('fs-364200', 1, 'whole'), ('fs-448393', 1, 'whole')],
+ bench=[('fs-278205', 1, 'whole'), ('mk-2585', 1, 'whole'),
+ ('fs-401803', 2, 'find', {'end': 5.0})]),
+ 'hover_tick': dict(query='a tiny soft tick', dur=(0.012, 0.15),
+ forge=['kit:ui_tick_01', 'kit:ui_tick_02', ('fs-376195', 2, 'find'), ('fs-329098', 2, 'find')],
+ bench=[('fs-401803', 3, 'find', {'start': 72.0}), ('fs-873306', 2, 'find', {'end': 5.2}),
+ ('fs-326470', 2, 'find')]),
+ 'pop': dict(query='a short round pop', dur=(0.04, 0.5),
+ forge=['kit:ui_pop_01', 'kit:ui_pop_02', 'kit:ui_pop_03', 'kit:ui_pop_04', ('mk-3000', 1, 'whole'),
+ ('fs-543200', 2, 'find')],
+ bench=[('fs-202230', 1, 'find'), ('fs-33369', 1, 'find'), ('fs-545201', 1, 'find'),
+ ('fs-573152', 2, 'find'), ('mk-3192', 1, 'whole'), ('mk-3070', 1, 'whole')]),
+ 'notify': dict(query='a short bright notification ding', dur=(0.4, 2.0),
+ forge=['kit:ui_notify_01', ('fs-681787', 1, 'whole'), ('fs-681789', 1, 'whole'),
+ ('fs-681792', 1, 'whole'), ('fs-549899', 1, 'whole')],
+ bench=[('fs-410636', 1, 'whole'), ('fs-410649', 1, 'whole'), ('fs-410664', 1, 'whole'),
+ ('fs-170237', 1, 'whole', {'end': 1.3}), ('vsco-Triangle6-Hit_v1_rr1', 1, 'whole')],
+ note='struck bells / glass / kalimba (forge) and celesta / glass / triangle (bench), rings capped '
+ 'at 2 s with a 0.25 s fade; no designed UI dings in the library.'),
+ 'shimmer': dict(query='a magical shimmering chime sweep', dur=(1.0, 6.0),
+ forge=['kit:shimmer_02', ('fs-271355', 1, 'find'), ('mk-2014', 1, 'whole'), ('fs-495327', 2, 'find')],
+ bench=['kit:shimmer_01', ('fs-257558', 1, 'find'), ('vsco-BellTree_Stroke1', 1, 'whole'),
+ ('vsco-glock_fx_up_chromatic_slow_01', 1, 'whole')],
+ note='kit shimmer_01 (bar chimes sweep) is bench: its author also recorded the celesta notify bench.'),
+ 'typing_key': dict(query='a single mechanical keyboard key press', dur=(0.06, 0.45),
+ forge=['kit:kb_thock_01', 'kit:kb_thock_02', 'kit:kb_thock_03', 'kit:kb_thock_04',
+ 'kit:kb_enter_01', 'kit:kb_space_01', 'kit:kb_click_01', 'kit:kb_click_02',
+ 'kit:kb_enter_02', ('fs-561697', 1, 'find'), ('fs-561699', 1, 'find'),
+ ('fs-709460', 2, 'find'), ('fs-665075', 2, 'find')],
+ bench=[('fs-378083', 1, 'find'), ('fs-378084', 1, 'find'), ('fs-378085', 1, 'find'),
+ ('fs-570754', 1, 'find'), ('fs-766639', 1, 'find'), ('fs-766640', 1, 'find'),
+ ('fs-609534', 3, 'find'), ('fs-515810', 1, 'find'), ('fs-815614', 2, 'find')]),
+ 'glitch': dict(query='a digital glitch stutter', dur=(0.1, 1.0), forge=[], bench=[],
+ note='empty: no glitch material, real or designed, in the licensed library. Benchmark made '
+ 'glitches on phrasings, AudioSet and artifacts only.'),
+ 'logo_sting': dict(query='a struck chime ringing out, a logo sting', dur=(1.0, 6.0),
+ forge=['kit:sting_tail_01', ('mk-3108', 1, 'whole')],
+ bench=[('vsco-Triangle6-Hit_v1_rr1', 1, 'whole'), ('fs-170237', 1, 'whole', {'end': 1.3})],
+ note='no real logo stings exist in the library (they are composed); the refs are the struck '
+ 'metal / glass rings a sting ends on: a tail-shape and tone reference only.'),
+}
+
+
+def sources():
+ with open(KIT + '/sources.json', encoding='utf-8') as fh:
+ return {s['id']: s for s in json.load(fh)}
+
+
+def kit_sounds():
+ with open(KIT + '/kit.json', encoding='utf-8') as fh:
+ k = json.load(fh)
+ return {s['name']: s for a in k['areas'] for s in a['sounds']}
+
+
+def family(src):
+ """sister takes stay together: Freesound by author, VSCO as one library, Mixkit by file"""
+ if src['id'].startswith('vsco-'):
+ return 'vsco'
+ return f"fs-author:{src['author']}" if src['id'].startswith('fs-') else src['id']
+
+
+def env_db(m, ms=5.0):
+ n = int(SR * ms / 1000)
+ k = max(1, len(m) // n)
+ return 20 * np.log10(np.sqrt((m[:k * n].reshape(k, n) ** 2).mean(1)) + 1e-12), n
+
+
+def cut_whole(m, dmax, rise=False):
+ """(a, b, capped) sample trims of the one sound in mono m (see the module docstring: whole / rise)"""
+ e, n = env_db(m)
+ p = int(np.argmax(e))
+ if rise:
+ j = p
+ while j > 0 and e[j - 1] > e[p] - 35 and (p - j + 1) * n < (dmax - 0.3) * SR:
+ j -= 1
+ return j * n, min(len(m), p * n + int(0.3 * SR)), False
+ on = int(np.nonzero(e >= e[p] - 30)[0][0])
+ a = max(0, on * n - int(0.005 * SR))
+ last = int(np.nonzero(e >= e[p] - 45)[0][-1])
+ b = min(len(m), (last + 1) * n + int(0.02 * SR))
+ if b - a > dmax * SR:
+ return a, a + int(dmax * SR), True
+ return a, b, False
+
+
+def load_cache():
+ try:
+ with open(CACHE, encoding='utf-8') as fh:
+ return json.load(fh)
+ except FileNotFoundError:
+ return {}
+
+
+def find_cached(cache, path, query, top, start=None, end=None):
+ key = f'{path}|{query}|{top}|{start}|{end}'
+ if key not in cache:
+ from earshot import api
+ r = api.find(path, query, top=top, start=start, end=end)
+ cache[key] = r['segments']
+ os.makedirs(os.path.dirname(CACHE), exist_ok=True)
+ with open(CACHE, 'w', encoding='utf-8') as fh:
+ json.dump(cache, fh)
+ return cache[key]
+
+
+def real_cut(s, t0, t1):
+ """find's 'head cut' / 'tail cut' flags, except at the edges of the searched range (a one-shot file that starts
+ on the attack cannot be trimmed earlier: its head is not cut)"""
+ return [c for c in (s['cut'] or '').split(', ') if c and not (c == 'head cut' and s['start'] <= t0 + 0.03)
+ and not (c == 'tail cut' and s['end'] >= t1 - 0.03)]
+
+
+def pick_found(segs, k, dur, taken, t0=0.0, t1=1e9, zmin=2.0):
+ """the best k clean segments inside the length range (module docstring: find); returns (picks, rejected)"""
+ loud = max((s['peak_dbfs'] for s in segs), default=-99)
+ picks, rej = [], []
+ for s in sorted(segs, key=lambda s: -s['score']):
+ why = ('edge cut' if real_cut(s, t0, t1) else f"length {s['duration']:.3f} s" if not dur[0] <= s['duration'] <= dur[1]
+ else f"z {s['z']:+.2f} < {zmin}" if s['z'] < zmin else f"{loud - s['peak_dbfs']:.0f} dB under the loudest"
+ if s['peak_dbfs'] < loud - 18 else
+ 'overlaps a kept trim' if any(s['start'] < b and s['end'] > a for a, b in taken) else '')
+ if why or len(picks) >= k:
+ rej.append({'start': s['start'], 'end': s['end'], 'z': s['z'], 'why': why or 'enough kept'})
+ continue
+ picks.append(s)
+ taken.append((s['start'], s['end']))
+ return picks, rej
+
+
+def export(x, a, b, path, fade_out):
+ """trim [a, b) samples of stereo x, 2 ms fade in, fade_out s out, peak to -1 dBFS; returns gain dB"""
+ import soundfile as sf
+ y = np.array(x[a:b], np.float64)
+ nf = min(len(y) // 4, int(0.002 * SR))
+ y[:nf] *= np.linspace(0, 1, nf)[:, None]
+ nf2 = max(1, min(len(y) // 3, int(fade_out * SR)))
+ y[-nf2:] *= np.linspace(1, 0, nf2)[:, None]
+ pk = float(np.abs(y).max()) + 1e-12
+ g = 10 ** (-1 / 20) / pk
+ os.makedirs(os.path.dirname(path), exist_ok=True)
+ sf.write(path, y * g, SR, subtype='PCM_24')
+ return round(20 * np.log10(g), 2)
+
+
+def build_cat(name, spec, S, K, cache, log=print):
+ out = {'query': spec['query'], 'dur': list(spec['dur']), 'note': spec.get('note', ''), 'forge': [], 'bench': [],
+ 'rejected': []}
+ for split in refsets.SPLITS:
+ taken = {}
+ for ent in spec[split]:
+ if isinstance(ent, str):
+ ks = K[ent[4:]]
+ src = next(s for s in S.values() if os.path.normpath(s['path']) == os.path.normpath(ks['source']['path']))
+ t0, t1 = ks['source']['trim_s']
+ x = audio.load(src['path'])
+ a, b = int(t0 * SR), min(len(x), int(t1 * SR))
+ capped = b - a > spec['dur'][1] * SR
+ b = a + int(spec['dur'][1] * SR) if capped else b
+ fo = 0.25 if capped else ks['processing'].get('fade_out_s') or 0.03
+ fname = f"kit-{ent[4:]}.wav"
+ how = {'how': 'kit trim' + (' (capped)' if capped else ''), 'kit_sound': ent[4:],
+ 'kit_match_query': ks['evidence'].get('match_query'), 'kit_match_z': ks['evidence'].get('match_z'),
+ 'kit_p_real': ks['evidence'].get('realness_p')}
+ cuts = [(a, b, fo, fname, how)]
+ taken.setdefault(src['id'], []).append((a / SR, b / SR))
+ else:
+ sid, k, mode = ent[:3]
+ opt = ent[3] if len(ent) > 3 else {}
+ src = S[sid]
+ x = audio.load(src['path'])
+ cuts = []
+ if mode == 'find':
+ segs = find_cached(cache, src['path'], spec['query'], max(8, 3 * k), opt.get('start'), opt.get('end'))
+ picks, rej = pick_found(segs, k, spec['dur'], taken.setdefault(sid, []), opt.get('start', 0.0),
+ opt.get('end', len(x) / SR), spec.get('zmin', 2.0))
+ out['rejected'] += [{'source_id': sid, 'split': split, **r} for r in rej]
+ for q, s in enumerate(picks):
+ cuts.append((int(s['start'] * SR), min(len(x), int(s['end'] * SR)),
+ min(0.03, s['duration'] / 4), f'{sid}_{q + 1}.wav',
+ {'how': 'find', 'find_z': s['z'], 'find_verdict': s['verdict'], 'find_kind': s['kind'],
+ 'head_db': s['head_db'], 'tail_db': s['tail_db'],
+ **({'edge': s['cut'] + ' at the file edge (kept)'} if s['cut'] else {})}))
+ if not picks:
+ log(f' {name}/{split}: nothing usable in {sid} ({len(segs)} candidates)')
+ else:
+ o0 = int(opt.get('start', 0) * SR)
+ o1 = int(opt['end'] * SR) if 'end' in opt else len(x)
+ a, b, capped = cut_whole(audio.to_mono(x[o0:o1]), spec['dur'][1], rise=mode == 'rise')
+ a, b = a + o0, b + o0
+ if (b - a) / SR < spec['dur'][0]:
+ out['rejected'].append({'source_id': sid, 'split': split, 'start': a / SR, 'end': b / SR,
+ 'why': f'{mode} cut {(b - a) / SR:.3f} s shorter than the range'})
+ else:
+ cuts.append((a, b, 0.25 if capped or mode == 'rise' else 0.03, f'{sid}_1.wav',
+ {'how': mode + (' (capped)' if capped else '')}))
+ for a, b, fo, fname, how in cuts:
+ path = f'{OUT}/{name}/{split}/{fname}'
+ g = export(x, a, b, path, fo)
+ y = audio.load(path, cache=False)
+ out[split].append({'file': path, 'source_id': src['id'], 'family': family(src), 'title': src['title'],
+ 'author': src['author'], 'licence': src['license'], 'page_url': src['page_url'],
+ 'source_path': src['path'], 'trim_s': [round(a / SR, 3), round(b / SR, 3)],
+ 'fade_in_s': 0.002, 'fade_out_s': round(fo, 3), 'gain_db': g,
+ 'evidence': {**how, 't40_ms': t40_ms(y), 'length_s': round(len(y) / SR, 3)}})
+ files = [r['file'] for s in refsets.SPLITS for r in out[s]]
+ if files:
+ from earshot import api
+ rk = {r['path'].replace(chr(92), '/'): r for r in api.rank(files, spec['query'])['results']}
+ for s in refsets.SPLITS:
+ for r in out[s]:
+ h = rk.get(r['file'])
+ if h:
+ r['evidence'].update(rank_z=h['z'], rank_verdict=h['verdict'], sounds_most_like=h['sounds_most_like'])
+ return out
+
+
+def registry(man):
+ reg = {'version': 1, 'built_by': 'eval/forge/build_refsets.py', 'root': OUT,
+ 'categories': {c: {s: [r['file'] for r in m[s]] for s in refsets.SPLITS} for c, m in man.items()}}
+ for c, m in man.items():
+ reg['categories'][c]['note'] = m['note']
+ fam = {r['file']: r['family'] for m in man.values() for s in refsets.SPLITS for r in m[s]}
+ return reg, refsets.check_split(reg, fam)
+
+
+def plan(S, K):
+ seen = {s: {} for s in refsets.SPLITS}
+ for c, spec in CATS.items():
+ for split in refsets.SPLITS:
+ ids = []
+ for ent in spec[split]:
+ if isinstance(ent, str):
+ p = K[ent[4:]]['source']['path']
+ sid = next(s['id'] for s in S.values() if os.path.normpath(s['path']) == os.path.normpath(p))
+ else:
+ sid = ent[0]
+ ids.append(sid)
+ seen[split].setdefault(family(S[sid]), set()).add(c)
+ print(f'{c:12s} {split:5s} {len(set(ids)):2d} sources: {" ".join(sorted(set(ids)))}')
+ bad = sorted(set(seen['forge']) & set(seen['bench']))
+ print('family leaks:', bad or 'none')
+ return bad
+
+
+def main(argv=None):
+ ap = argparse.ArgumentParser(description=__doc__.split('\n')[0])
+ ap.add_argument('--cats', nargs='*', help='only these categories (merged into the existing manifest)')
+ ap.add_argument('--plan', action='store_true', help='print the source plan and the leak check, no audio work')
+ ap.add_argument('--registry', help='registry path (default earshot.forge.refsets.registry_path())')
+ a = ap.parse_args(argv)
+ S, K = sources(), kit_sounds()
+ if plan(S, K) or a.plan:
+ return
+ mpath = OUT + '/manifest.json'
+ man = {}
+ if os.path.exists(mpath):
+ with open(mpath, encoding='utf-8') as fh:
+ man = json.load(fh)
+ cache = load_cache()
+ for c in a.cats or list(CATS):
+ print(f'{c} ...', flush=True)
+ for s in refsets.SPLITS:
+ d = f'{OUT}/{c}/{s}'
+ if os.path.isdir(d):
+ for f in os.listdir(d):
+ os.remove(os.path.join(d, f))
+ man[c] = build_cat(c, CATS[c], S, K, cache)
+ print(f" forge {len(man[c]['forge'])}, bench {len(man[c]['bench'])}, rejected {len(man[c]['rejected'])}")
+ with open(mpath, 'w', encoding='utf-8') as fh:
+ json.dump(man, fh, indent=1)
+ man = {c: man[c] for c in CATS if c in man}
+ with open(mpath, 'w', encoding='utf-8') as fh:
+ json.dump(man, fh, indent=1)
+ reg, bad = registry(man)
+ if bad:
+ raise SystemExit('split leak, registry not written: ' + '; '.join(bad))
+ print('registry:', refsets.write(reg, a.registry))
+
+
+if __name__ == '__main__':
+ main()
diff --git a/eval/forge/frozen_tone.py b/eval/forge/frozen_tone.py
new file mode 100644
index 0000000..8480430
--- /dev/null
+++ b/eval/forge/frozen_tone.py
@@ -0,0 +1,180 @@
+"""FROZEN copy of earshot/forge/tone.py (profile, shares, t40_ms, ToneRef) as of branch forge-v2 round 0 (git HEAD
+108763c), so the forge benchmark measures the same thing every round while the forge's own tone.py evolves.
+
+ from frozen_tone import ToneRef, profile, t40_ms (eval/forge/bench.py imports it from this folder)
+
+Do not edit: a change here changes every benchmark number. Add new measurements in bench.py instead (and record them in
+OVERNIGHT.md). The original docstring follows.
+
+Model-free ears for the forge: tonal balance and tail, measured straight from the samples.
+
+ from earshot.forge.tone import ToneRef
+ tr = ToneRef(['real_01.wav', 'real_02.wav', ...]) # files or 48 kHz arrays
+ m = tr.check(x)
+ m['tone_db'], m['tone_ratio'], m['words'] # e.g. 'tone 11.2 dB from the refs (they sit 3.1-6.0 apart):
+ # +16 dB at 500-630 Hz (above every ref), -14 dB at 10 kHz (below ...)'
+
+Why this exists: earshot's model signals saturate. After a search tunes a sound against CLAP (match z, typicality) and
+the realness head, the sound lands inside or above the real recordings on every model check, including held-out
+phrasings, the AudioSet tagger and independent real sets (demo/forge-test, the judge's re-scoring). The models can no
+longer tell made from real, but a 1/3-octave measurement still can (a key press 8-13 dB off the real keys' balance
+while the real keys sit 3-6 dB from each other). These numbers cannot be pleased by a room and a mic noise floor.
+
+Measures (mono sum, the whole file):
+ profile power per 1/3-octave band, 31.5 Hz - 16 kHz (28 bands), in dB relative to the file's total power
+ (so level never decides), floored at -60 dB
+ tone_db RMS over bands of (profile - the refs' median profile)
+ tone_ref the refs' own distances, each to the median of the OTHER refs (leave-one-out): what "real" spreads to
+ tone_ratio tone_db / median(tone_ref): 1 = as far from the median as a typical real recording
+ outside bands where the sound is beyond the refs' min / max by more than `margin` dB (the words name the worst)
+ t40_ms time from the peak until the 5 ms RMS envelope first falls 40 dB under it (tail length), with the refs'
+ range; length_s the file length
+ shares energy share of 5 broad regions (<150 Hz, 150 Hz-1 kHz, 1-4, 4-12, >12 kHz) in dB of the total, with the
+ refs' median share; region_penalty() counts the excess beyond median + 6 dB (a low puff, a bright hiss)
+"""
+import numpy as np
+
+SR = 48000
+CEN = 1000 * 2 ** (np.arange(-15, 13) / 3) # 31.5 Hz .. 15.9 kHz
+EDGE = 2 ** (1 / 6)
+REGIONS = {'<150': (0, 150), '150-1k': (150, 1000), '1-4k': (1000, 4000), '4-12k': (4000, 12000), '>12k': (12000, 24000)}
+
+
+def _mono(x):
+ x = np.asarray(x, np.float64)
+ return x.mean(axis=1) if x.ndim == 2 else x
+
+
+def _load(x):
+ if isinstance(x, str):
+ from earshot import audio
+ return audio.load(x)
+ return x
+
+
+def label(c):
+ return f'{c / 1000:.3g} kHz' if c >= 1000 else f'{c:.0f} Hz'
+
+
+def profile(x):
+ """1/3-octave power profile in dB relative to the total power (28 bands, floored at -60)"""
+ m = _mono(x)
+ F = np.abs(np.fft.rfft(m)) ** 2
+ f = np.fft.rfftfreq(len(m), 1 / SR)
+ p = np.array([F[(f >= c / EDGE) & (f < c * EDGE)].sum() for c in CEN]) + 1e-20
+ return np.maximum(10 * np.log10(p / p.sum()), -60.0)
+
+
+def shares(x):
+ """energy share of each broad region (REGIONS) in dB of the total"""
+ m = _mono(x)
+ F = np.abs(np.fft.rfft(m)) ** 2
+ f = np.fft.rfftfreq(len(m), 1 / SR)
+ tot = F.sum() + 1e-20
+ return {k: round(float(10 * np.log10(F[(f >= a) & (f < b)].sum() / tot + 1e-12)), 1) for k, (a, b) in REGIONS.items()}
+
+
+def t40_ms(x, db=40.0, ms=5.0):
+ """ms from the peak of the 5 ms RMS envelope until it first falls `db` under it"""
+ m = _mono(x)
+ n = int(SR * ms / 1000)
+ k = len(m) // n
+ if k < 2:
+ return 0.0
+ e = 20 * np.log10(np.sqrt((m[:k * n].reshape(k, n) ** 2).mean(1)) + 1e-12)
+ i = int(np.argmax(e))
+ after = e[i:] - e[i]
+ idx = np.nonzero(after < -db)[0]
+ return float((idx[0] if len(idx) else len(after)) * ms)
+
+
+class ToneRef:
+ """the real references' tonal balance and tail, and a check of any sound against them"""
+
+ def __init__(self, refs, margin=3.0):
+ xs = [_load(r) for r in refs]
+ self.n = len(xs)
+ self.margin = margin
+ self.P = np.stack([profile(x) for x in xs]) if xs else np.zeros((0, len(CEN)))
+ self.med = np.median(self.P, 0) if xs else None
+ self.lo, self.hi = (self.P.min(0), self.P.max(0)) if xs else (None, None)
+ self.t40 = np.array([t40_ms(x) for x in xs])
+ self.S = [shares(x) for x in xs]
+ self.len_s = np.array([len(_mono(x)) / SR for x in xs])
+ self.ref_dist = np.array([self._loo(i) for i in range(self.n)]) if self.n >= 3 else np.zeros(0)
+
+ def _loo(self, i):
+ keep = [j for j in range(self.n) if j != i]
+ return float(np.sqrt(np.mean((self.P[i] - np.median(self.P[keep], 0)) ** 2)))
+
+ @property
+ def spread(self):
+ """(median, max) of the refs' leave-one-out distances, or None with fewer than 3 refs"""
+ if not len(self.ref_dist):
+ return None
+ return float(np.median(self.ref_dist)), float(self.ref_dist.max())
+
+ def check(self, x, exclude=None):
+ """measure one sound against the refs. exclude = index of a ref to leave out (scoring a ref fairly)"""
+ x = _load(x)
+ if self.n == 0:
+ return {}
+ keep = [j for j in range(self.n) if j != exclude]
+ P = self.P[keep]
+ med, lo, hi = np.median(P, 0), P.min(0), P.max(0)
+ p = profile(x)
+ dev = p - med
+ dist = float(np.sqrt(np.mean(dev ** 2)))
+ sp = self.ref_dist[keep] if len(self.ref_dist) else np.zeros(0)
+ ratio = dist / max(float(np.median(sp)), 0.5) if len(sp) >= 2 else None
+ over = np.maximum(p - hi - self.margin, 0) - np.maximum(lo - p - self.margin, 0)
+ outside = [(label(c), round(float(d), 1), 'above' if o > 0 else 'below')
+ for c, d, o in zip(CEN, dev, over) if o != 0]
+ t40 = t40_ms(x)
+ t40r = self.t40[keep]
+ out = {'tone_db': round(dist, 2), 'tone_ratio': None if ratio is None else round(ratio, 2),
+ 'tone_ref': [round(float(v), 2) for v in sp], 'band_dev': [round(float(v), 1) for v in dev],
+ 'outside': outside, 't40_ms': t40, 't40_ref': (float(t40r.min()), float(t40r.max())),
+ 'length_s': round(len(_mono(x)) / SR, 3), 'shares': shares(x),
+ 'shares_ref_med': {k: round(float(np.median([self.S[j][k] for j in keep])), 1) for k in REGIONS},
+ 'length_ref': (round(float(self.len_s[keep].min()), 3), round(float(self.len_s[keep].max()), 3))}
+ out['words'] = self.words(out)
+ return out
+
+ @staticmethod
+ def words(m):
+ sp = m['tone_ref']
+ s = f"tone {m['tone_db']:.1f} dB from the refs' median"
+ if sp:
+ s += f" (the refs sit {min(sp):.1f}-{max(sp):.1f} apart)"
+ worst = sorted(m['outside'], key=lambda o: -abs(o[1]))[:3]
+ if worst:
+ s += ': ' + ', '.join(f'{d:+.0f} dB at {b} ({w} every ref)' for b, d, w in worst)
+ a, b = m['t40_ref']
+ if m['t40_ms'] > 1.5 * b + 10 or m['t40_ms'] < 0.6 * a - 5:
+ s += f"; tail t40 {m['t40_ms']:.0f} ms (refs {a:.0f}-{b:.0f})"
+ loud = [f"{k} {m['shares'][k] - m['shares_ref_med'][k]:+.0f} dB" for k in REGIONS
+ if abs(m['shares'][k] - m['shares_ref_med'][k]) > 9]
+ if loud:
+ s += '; energy share vs the refs: ' + ', '.join(loud)
+ lo, hi = m['length_ref']
+ if m['length_s'] > 1.3 * hi:
+ s += f"; {m['length_s']:.2f} s long (refs {lo:.2f}-{hi:.2f})"
+ return s
+
+ @staticmethod
+ def region_penalty(m, regions=('<150', '4-12k', '>12k'), margin=6.0, scale=6.0):
+ """model-free check of broad regions: 1 per `scale` dB that a region's energy share exceeds the refs' median
+ share by more than `margin` dB (a low puff or a bright hiss / tick on a mid-band sound)"""
+ return sum(max(0.0, m['shares'][k] - m['shares_ref_med'][k] - margin) / scale for k in regions)
+
+ def penalty(self, m):
+ """0 while the sound is no farther from the refs' median than a typical real recording, then linear in the
+ excess (tone_ratio - 1), plus a share of the tail excess beyond the refs' longest t40"""
+ if not m or m.get('tone_ratio') is None:
+ return 0.0
+ pen = max(0.0, m['tone_ratio'] - 1.0)
+ a, b = m['t40_ref']
+ if b > 0 and m['t40_ms'] > b:
+ pen += 0.5 * min(2.0, (m['t40_ms'] - b) / b)
+ return pen
diff --git a/eval/forge/results_real.json b/eval/forge/results_real.json
new file mode 100644
index 0000000..19d9ef8
--- /dev/null
+++ b/eval/forge/results_real.json
@@ -0,0 +1,16875 @@
+{
+ "tag": "real",
+ "made_from": null,
+ "registry": "E:\\earshot\\.cache\\refsets.json",
+ "created": "2026-10-16 04:07",
+ "phrases": {
+ "click": [
+ "a crisp tiny plastic click",
+ "a short clean click of a button being pressed"
+ ],
+ "downlifter": [
+ "a descending whoosh that fades out",
+ "a falling pitch sweep winding down"
+ ],
+ "glitch": [
+ "a digital glitch with stuttering noise",
+ "a broken electronic signal stutter"
+ ],
+ "hover_tick": [
+ "a very short faint tick",
+ "a tiny dry click"
+ ],
+ "impact": [
+ "a huge low thud for a movie trailer",
+ "a massive hit with a rumbling tail"
+ ],
+ "logo_sting": [
+ "a short musical logo with a ringing finish",
+ "a bright struck bell ringing out"
+ ],
+ "notify": [
+ "a pleasant chime alert sound",
+ "a single bell ding on a phone"
+ ],
+ "pop": [
+ "a popping sound like a bubble bursting",
+ "a quick plop"
+ ],
+ "riser": [
+ "a tension build-up getting louder and louder",
+ "a whooshing crescendo that swells to a climax"
+ ],
+ "shimmer": [
+ "sparkling twinkling chimes",
+ "a glittering magical sound"
+ ],
+ "sub_drop": [
+ "a low frequency boom that drops in pitch",
+ "a deep rumbling bass drop"
+ ],
+ "swipe": [
+ "fabric rustling in one quick swipe",
+ "a sleeve brushing past quickly"
+ ],
+ "swish": [
+ "a thin airy swish",
+ "a light quick swoosh of a thrown object"
+ ],
+ "tap": [
+ "a fingertip tapping on glass",
+ "a light tap of a finger on a phone screen"
+ ],
+ "toggle": [
+ "a light switch clicking on",
+ "a small switch being flipped"
+ ],
+ "typing_key": [
+ "one key being pressed on a computer keyboard",
+ "a single keystroke on a mechanical keyboard"
+ ],
+ "whoosh_big": [
+ "a heavy swooping rush of air",
+ "a big slow swoosh in a movie trailer"
+ ],
+ "whoosh_fast": [
+ "a quick swoosh as something flies past",
+ "a short sharp swoosh of air"
+ ]
+ },
+ "confusions": {
+ "click": [
+ "a wooden stick hit",
+ "a finger snap",
+ "a drop of water",
+ "a stapler"
+ ],
+ "downlifter": [
+ "a siren",
+ "a car passing by",
+ "a falling bomb whistle",
+ "wind"
+ ],
+ "glitch": [
+ "radio static",
+ "a modem dialing",
+ "a scratched record",
+ "an electric buzz"
+ ],
+ "hover_tick": [
+ "a clock ticking",
+ "a drop of water",
+ "a typewriter",
+ "a cricket chirp"
+ ],
+ "impact": [
+ "a gunshot",
+ "a door being slammed",
+ "a bass drum",
+ "distant fireworks"
+ ],
+ "logo_sting": [
+ "a church bell",
+ "a doorbell",
+ "a gong",
+ "a cymbal crash"
+ ],
+ "notify": [
+ "a doorbell",
+ "a microwave beep",
+ "a church bell",
+ "a cash register"
+ ],
+ "pop": [
+ "a hand clap",
+ "a gunshot",
+ "a kiss",
+ "a tongue click"
+ ],
+ "riser": [
+ "a jet engine starting",
+ "a siren",
+ "steady rain",
+ "an orchestra tuning"
+ ],
+ "shimmer": [
+ "a harp glissando",
+ "breaking glass",
+ "a music box",
+ "a sleigh bell jingle"
+ ],
+ "sub_drop": [
+ "an earthquake rumble",
+ "a kick drum",
+ "a truck engine idling",
+ "thunder in the distance"
+ ],
+ "swipe": [
+ "paper crumpling",
+ "a broom sweeping",
+ "a gust of wind",
+ "a zipper"
+ ],
+ "swish": [
+ "a short hiss",
+ "a sword being drawn",
+ "a page being turned",
+ "a bird flying past"
+ ],
+ "tap": [
+ "a knock on a door",
+ "a raindrop",
+ "a keyboard key",
+ "a ballpoint pen click"
+ ],
+ "toggle": [
+ "a door latch",
+ "a stapler",
+ "a seatbelt buckle",
+ "a ball bouncing"
+ ],
+ "typing_key": [
+ "a ballpoint pen click",
+ "a stapler",
+ "a light switch",
+ "a dripping tap"
+ ],
+ "whoosh_big": [
+ "a strong wind storm",
+ "an airplane taking off",
+ "ocean waves crashing",
+ "a car driving past"
+ ],
+ "whoosh_fast": [
+ "a steam hiss",
+ "a car driving past",
+ "tearing paper",
+ "a bird flapping its wings"
+ ]
+ },
+ "as_labels": {
+ "click": [
+ "Clicking",
+ "Tick",
+ "Tap"
+ ],
+ "downlifter": [
+ "Whoosh, swoosh, swish"
+ ],
+ "glitch": [
+ "Static",
+ "Noise",
+ "Beep, bleep",
+ "White noise",
+ "Crackle"
+ ],
+ "hover_tick": [
+ "Tick",
+ "Clicking",
+ "Tick-tock"
+ ],
+ "impact": [
+ "Boom",
+ "Thump, thud",
+ "Explosion",
+ "Bang",
+ "Bass drum"
+ ],
+ "logo_sting": [
+ "Chime",
+ "Bell",
+ "Ding",
+ "Jingle, tinkle",
+ "Gong",
+ "Cymbal",
+ "Glockenspiel"
+ ],
+ "notify": [
+ "Ding",
+ "Chime",
+ "Bell",
+ "Glockenspiel",
+ "Jingle, tinkle",
+ "Ding-dong",
+ "Ringtone",
+ "Marimba, xylophone",
+ "Vibraphone"
+ ],
+ "pop": [
+ "Burst, pop",
+ "Plop"
+ ],
+ "riser": [
+ "Whoosh, swoosh, swish",
+ "Cymbal",
+ "Drum roll"
+ ],
+ "shimmer": [
+ "Chime",
+ "Wind chime",
+ "Jingle, tinkle",
+ "Glockenspiel",
+ "Bell",
+ "Jingle bell"
+ ],
+ "sub_drop": [
+ "Boom",
+ "Rumble",
+ "Thump, thud"
+ ],
+ "swipe": [
+ "Whoosh, swoosh, swish",
+ "Rustle"
+ ],
+ "swish": [
+ "Whoosh, swoosh, swish",
+ "Whip"
+ ],
+ "tap": [
+ "Tap",
+ "Knock",
+ "Clicking",
+ "Tick"
+ ],
+ "toggle": [
+ "Clicking",
+ "Mechanisms",
+ "Tick"
+ ],
+ "typing_key": [
+ "Computer keyboard",
+ "Typing",
+ "Typewriter",
+ "Clicking"
+ ],
+ "whoosh_big": [
+ "Whoosh, swoosh, swish"
+ ],
+ "whoosh_fast": [
+ "Whoosh, swoosh, swish",
+ "Whip"
+ ]
+ },
+ "categories": {
+ "click": {
+ "n_bench": 7,
+ "n_forge": 8,
+ "note": "",
+ "bench_refs": [
+ "E:/earshot/out/refs/click/bench/mk-1124_1.wav",
+ "E:/earshot/out/refs/click/bench/mk-2364_1.wav",
+ "E:/earshot/out/refs/click/bench/fs-610524_1.wav",
+ "E:/earshot/out/refs/click/bench/fs-461799_1.wav",
+ "E:/earshot/out/refs/click/bench/fs-461799_2.wav",
+ "E:/earshot/out/refs/click/bench/fs-431419_1.wav",
+ "E:/earshot/out/refs/click/bench/fs-431419_2.wav"
+ ],
+ "made": [],
+ "diversity": {
+ "n": 0
+ },
+ "real_cross": [
+ {
+ "file": "E:/earshot/out/refs/click/forge/kit-ui_click_01.wav",
+ "length_s": 0.096,
+ "typ_z": -2.17,
+ "nn_cos": 0.621,
+ "tone_db": 20.26,
+ "tone_ratio": 2.06,
+ "n_outside": 13,
+ "outside": [
+ [
+ "31 Hz",
+ 44.1,
+ "above"
+ ],
+ [
+ "4 kHz",
+ -32.4,
+ "below"
+ ],
+ [
+ "3.17 kHz",
+ -30.5,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 60.0,
+ "active_s": 0.082,
+ "onset_s": 0.008,
+ "env_oct": 0.208,
+ "curve_db": 9.91,
+ "t40_ratio": 1.333,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.155,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.071
+ },
+ "phrase_z": 2.54,
+ "phrase_min": 2.14,
+ "margin": 1.05,
+ "wins": 0.5,
+ "best_confusion": "a drop of water",
+ "as_top": [
+ [
+ "Speech",
+ 0.125
+ ],
+ [
+ "Vehicle",
+ 0.101
+ ],
+ [
+ "Music",
+ 0.094
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -43.6,
+ "stereo_corr": 1.0,
+ "darken": 1.51,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -52.9,
+ "first_sample": 0.0,
+ "tail_db": -39.4,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/forge/kit-ui_click_02.wav",
+ "length_s": 0.118,
+ "typ_z": 0.04,
+ "nn_cos": 0.72,
+ "tone_db": 7.78,
+ "tone_ratio": 0.79,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 1.0,
+ "t40_ms": 110.0,
+ "active_s": 0.107,
+ "onset_s": 0.008,
+ "env_oct": 0.627,
+ "curve_db": 14.28,
+ "t40_ratio": 2.444,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.507,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.071
+ },
+ "phrase_z": 2.37,
+ "phrase_min": 2.33,
+ "margin": 0.07,
+ "wins": 0.0,
+ "best_confusion": "a finger snap",
+ "as_top": [
+ [
+ "Speech",
+ 0.119
+ ],
+ [
+ "Slap, smack",
+ 0.047
+ ],
+ [
+ "Music",
+ 0.039
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.012,
+ "floor_db": -35.1,
+ "stereo_corr": 0.916,
+ "darken": 0.23,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -64.3,
+ "first_sample": 0.0,
+ "tail_db": -38.5,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/forge/kit-ui_click_03.wav",
+ "length_s": 0.159,
+ "typ_z": 0.94,
+ "nn_cos": 0.755,
+ "tone_db": 21.05,
+ "tone_ratio": 2.14,
+ "n_outside": 15,
+ "outside": [
+ [
+ "630 Hz",
+ -35.9,
+ "below"
+ ],
+ [
+ "500 Hz",
+ -33.4,
+ "below"
+ ],
+ [
+ "99 Hz",
+ -33.3,
+ "below"
+ ]
+ ],
+ "attack_ms": 5.0,
+ "t40_ms": 30.0,
+ "active_s": 0.032,
+ "onset_s": 0.008,
+ "env_oct": 1.352,
+ "curve_db": 10.15,
+ "t40_ratio": 0.667,
+ "attack_ratio": 5.0,
+ "active_ratio": 0.451,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.071
+ },
+ "phrase_z": 2.97,
+ "phrase_min": 2.9,
+ "margin": 0.25,
+ "wins": 1.0,
+ "best_confusion": "a finger snap",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.107
+ ],
+ [
+ "Slap, smack",
+ 0.095
+ ],
+ [
+ "Music",
+ 0.045
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.004,
+ "floor_db": -58.0,
+ "stereo_corr": 0.973,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -112.3,
+ "first_sample": 0.0,
+ "tail_db": -59.4,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/forge/kit-ui_click_04.wav",
+ "length_s": 0.072,
+ "typ_z": 0.29,
+ "nn_cos": 0.702,
+ "tone_db": 12.86,
+ "tone_ratio": 1.31,
+ "n_outside": 7,
+ "outside": [
+ [
+ "16 kHz",
+ -37.5,
+ "below"
+ ],
+ [
+ "12.7 kHz",
+ -25.9,
+ "below"
+ ],
+ [
+ "39 Hz",
+ 20.5,
+ "above"
+ ]
+ ],
+ "attack_ms": 3.0,
+ "t40_ms": 55.0,
+ "active_s": 0.058,
+ "onset_s": 0.007,
+ "env_oct": 0.722,
+ "curve_db": 9.93,
+ "t40_ratio": 1.222,
+ "attack_ratio": 3.0,
+ "active_ratio": 0.817,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.071
+ },
+ "phrase_z": 2.77,
+ "phrase_min": 2.41,
+ "margin": 0.52,
+ "wins": 0.5,
+ "best_confusion": "a wooden stick hit",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.069
+ ],
+ [
+ "Music",
+ 0.024
+ ],
+ [
+ "Speech",
+ 0.024
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -34.0,
+ "stereo_corr": 1.0,
+ "darken": -0.63,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -54.2,
+ "first_sample": 0.0,
+ "tail_db": -37.4,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/forge/fs-167829_1.wav",
+ "length_s": 0.072,
+ "typ_z": 1.72,
+ "nn_cos": 0.807,
+ "tone_db": 6.17,
+ "tone_ratio": 0.63,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 1.0,
+ "t40_ms": 55.0,
+ "active_s": 0.055,
+ "onset_s": 0.005,
+ "env_oct": 0.219,
+ "curve_db": 8.44,
+ "t40_ratio": 1.222,
+ "attack_ratio": 1.0,
+ "active_ratio": 0.775,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.071
+ },
+ "phrase_z": 2.7,
+ "phrase_min": 2.55,
+ "margin": -0.27,
+ "wins": 0.0,
+ "best_confusion": "a finger snap",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.062
+ ],
+ [
+ "Speech",
+ 0.028
+ ],
+ [
+ "Music",
+ 0.028
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -41.5,
+ "stereo_corr": 0.919,
+ "darken": 0.02,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -69.7,
+ "first_sample": 0.0,
+ "tail_db": -48.4,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/forge/fs-167829_2.wav",
+ "length_s": 0.074,
+ "typ_z": 1.5,
+ "nn_cos": 0.769,
+ "tone_db": 7.53,
+ "tone_ratio": 0.76,
+ "n_outside": 3,
+ "outside": [
+ [
+ "62 Hz",
+ 17.4,
+ "above"
+ ],
+ [
+ "50 Hz",
+ 13.9,
+ "above"
+ ],
+ [
+ "39 Hz",
+ 13.5,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 60.0,
+ "active_s": 0.06,
+ "onset_s": 0.005,
+ "env_oct": 0.219,
+ "curve_db": 8.61,
+ "t40_ratio": 1.333,
+ "attack_ratio": 1.0,
+ "active_ratio": 0.845,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.071
+ },
+ "phrase_z": 2.77,
+ "phrase_min": 2.65,
+ "margin": 0.92,
+ "wins": 1.0,
+ "best_confusion": "a wooden stick hit",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.068
+ ],
+ [
+ "Speech",
+ 0.03
+ ],
+ [
+ "Music",
+ 0.024
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -42.1,
+ "stereo_corr": 0.934,
+ "darken": -0.12,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -82.5,
+ "first_sample": 0.0,
+ "tail_db": -45.8,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/forge/fs-383863_1.wav",
+ "length_s": 0.165,
+ "typ_z": -0.36,
+ "nn_cos": 0.673,
+ "tone_db": 11.27,
+ "tone_ratio": 1.14,
+ "n_outside": 3,
+ "outside": [
+ [
+ "16 kHz",
+ -37.5,
+ "below"
+ ],
+ [
+ "315 Hz",
+ 15.0,
+ "above"
+ ],
+ [
+ "1.26 kHz",
+ 12.7,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 70.0,
+ "active_s": 0.072,
+ "onset_s": 0.073,
+ "env_oct": 0.219,
+ "curve_db": 7.33,
+ "t40_ratio": 1.556,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.014,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.071
+ },
+ "phrase_z": 2.88,
+ "phrase_min": 2.62,
+ "margin": 1.11,
+ "wins": 1.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.113
+ ],
+ [
+ "Sound effect",
+ 0.052
+ ],
+ [
+ "Tap",
+ 0.037
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.037,
+ "floor_db": -56.4,
+ "stereo_corr": 1.0,
+ "darken": -1.02,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -85.2,
+ "first_sample": 0.0,
+ "tail_db": -47.4,
+ "lead_s": 0.057,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/forge/fs-383863_2.wav",
+ "length_s": 0.18,
+ "typ_z": -1.1,
+ "nn_cos": 0.649,
+ "tone_db": 11.39,
+ "tone_ratio": 1.16,
+ "n_outside": 4,
+ "outside": [
+ [
+ "16 kHz",
+ -37.5,
+ "below"
+ ],
+ [
+ "2.52 kHz",
+ -15.1,
+ "below"
+ ],
+ [
+ "3.17 kHz",
+ -11.8,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 70.0,
+ "active_s": 0.071,
+ "onset_s": 0.089,
+ "env_oct": 0.212,
+ "curve_db": 10.66,
+ "t40_ratio": 1.556,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.0,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.071
+ },
+ "phrase_z": 2.83,
+ "phrase_min": 2.55,
+ "margin": 0.53,
+ "wins": 1.0,
+ "best_confusion": "a wooden stick hit",
+ "as_top": [
+ [
+ "Chopping (food)",
+ 0.649
+ ],
+ [
+ "Finger snapping",
+ 0.025
+ ],
+ [
+ "Chop",
+ 0.024
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.004,
+ "floor_db": -49.8,
+ "stereo_corr": 1.0,
+ "darken": -0.61,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -72.5,
+ "first_sample": 0.0,
+ "tail_db": -56.7,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_loo": [
+ {
+ "file": "E:/earshot/out/refs/click/bench/mk-1124_1.wav",
+ "length_s": 0.055,
+ "typ_z": 0.58,
+ "nn_cos": 0.699,
+ "tone_db": 14.31,
+ "tone_ratio": 1.5,
+ "n_outside": 12,
+ "outside": [
+ [
+ "12.7 kHz",
+ -25.6,
+ "below"
+ ],
+ [
+ "500 Hz",
+ -21.4,
+ "below"
+ ],
+ [
+ "50 Hz",
+ -20.2,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 25.0,
+ "active_s": 0.022,
+ "onset_s": 0.008,
+ "env_oct": 0.972,
+ "curve_db": 13.19,
+ "t40_ratio": 0.455,
+ "attack_ratio": 1.0,
+ "active_ratio": 0.291,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 55.0,
+ "active_s": 0.075
+ },
+ "phrase_z": 2.67,
+ "phrase_min": 2.65,
+ "margin": 0.63,
+ "wins": 1.0,
+ "best_confusion": "a drop of water",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.063
+ ],
+ [
+ "Music",
+ 0.045
+ ],
+ [
+ "Speech",
+ 0.032
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -68.2,
+ "stereo_corr": 1.0,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -94.4,
+ "first_sample": 0.0,
+ "tail_db": -71.2,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/bench/mk-2364_1.wav",
+ "length_s": 0.21,
+ "typ_z": -2.5,
+ "nn_cos": 0.699,
+ "tone_db": 11.83,
+ "tone_ratio": 1.24,
+ "n_outside": 8,
+ "outside": [
+ [
+ "31 Hz",
+ 30.5,
+ "above"
+ ],
+ [
+ "794 Hz",
+ 21.7,
+ "above"
+ ],
+ [
+ "630 Hz",
+ 20.9,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.08,
+ "onset_s": 0.005,
+ "env_oct": 0.205,
+ "curve_db": 17.22,
+ "t40_ratio": 0.857,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.311,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 52.5,
+ "active_s": 0.061
+ },
+ "phrase_z": 2.4,
+ "phrase_min": 2.21,
+ "margin": -0.06,
+ "wins": 0.5,
+ "best_confusion": "a drop of water",
+ "as_top": [
+ [
+ "Plop",
+ 0.667
+ ],
+ [
+ "Music",
+ 0.045
+ ],
+ [
+ "Sound effect",
+ 0.042
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.001,
+ "floor_db": -47.3,
+ "stereo_corr": 1.0,
+ "darken": -2.12,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -53.9,
+ "first_sample": 0.0,
+ "tail_db": -59.7,
+ "lead_s": 0.004,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/bench/fs-610524_1.wav",
+ "length_s": 0.083,
+ "typ_z": -0.31,
+ "nn_cos": 0.712,
+ "tone_db": 11.51,
+ "tone_ratio": 1.21,
+ "n_outside": 9,
+ "outside": [
+ [
+ "99 Hz",
+ 23.1,
+ "above"
+ ],
+ [
+ "198 Hz",
+ 22.8,
+ "above"
+ ],
+ [
+ "125 Hz",
+ 20.4,
+ "above"
+ ]
+ ],
+ "attack_ms": 16.0,
+ "t40_ms": 35.0,
+ "active_s": 0.051,
+ "onset_s": 0.014,
+ "env_oct": 1.739,
+ "curve_db": 12.77,
+ "t40_ratio": 0.636,
+ "attack_ratio": 16.0,
+ "active_ratio": 0.675,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 55.0,
+ "active_s": 0.075
+ },
+ "phrase_z": 2.92,
+ "phrase_min": 2.89,
+ "margin": 1.38,
+ "wins": 1.0,
+ "best_confusion": "a finger snap",
+ "as_top": [
+ [
+ "Speech",
+ 0.053
+ ],
+ [
+ "Music",
+ 0.053
+ ],
+ [
+ "Sound effect",
+ 0.04
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -44.8,
+ "stereo_corr": -0.129,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -61.6,
+ "first_sample": 0.0,
+ "tail_db": -53.6,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/bench/fs-461799_1.wav",
+ "length_s": 0.305,
+ "typ_z": -0.2,
+ "nn_cos": 0.68,
+ "tone_db": 6.79,
+ "tone_ratio": 0.64,
+ "n_outside": 1,
+ "outside": [
+ [
+ "2 kHz",
+ 7.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 149.0,
+ "t40_ms": 110.0,
+ "active_s": 0.255,
+ "onset_s": 0.03,
+ "env_oct": 3.552,
+ "curve_db": 26.17,
+ "t40_ratio": 2.588,
+ "attack_ratio": 149.0,
+ "active_ratio": 4.18,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 42.5,
+ "active_s": 0.061
+ },
+ "phrase_z": 2.34,
+ "phrase_min": 2.33,
+ "margin": -0.44,
+ "wins": 0.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.578
+ ],
+ [
+ "Clapping",
+ 0.447
+ ],
+ [
+ "Speech",
+ 0.045
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.024,
+ "floor_db": -38.9,
+ "stereo_corr": 0.824,
+ "darken": -0.35,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -74.6,
+ "first_sample": 0.0,
+ "tail_db": -48.8,
+ "lead_s": 0.005,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/bench/fs-461799_2.wav",
+ "length_s": 0.109,
+ "typ_z": -0.1,
+ "nn_cos": 0.681,
+ "tone_db": 8.56,
+ "tone_ratio": 0.8,
+ "n_outside": 2,
+ "outside": [
+ [
+ "125 Hz",
+ -21.5,
+ "below"
+ ],
+ [
+ "1.59 kHz",
+ 11.0,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 70.0,
+ "active_s": 0.071,
+ "onset_s": 0.019,
+ "env_oct": 0.279,
+ "curve_db": 12.26,
+ "t40_ratio": 1.647,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.084,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 42.5,
+ "active_s": 0.066
+ },
+ "phrase_z": 2.51,
+ "phrase_min": 2.48,
+ "margin": -0.19,
+ "wins": 0.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Speech",
+ 0.114
+ ],
+ [
+ "Music",
+ 0.034
+ ],
+ [
+ "Single-lens reflex camera",
+ 0.028
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.003,
+ "floor_db": -40.0,
+ "stereo_corr": 0.948,
+ "darken": 0.13,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -90.1,
+ "first_sample": 0.0,
+ "tail_db": -50.3,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/bench/fs-431419_1.wav",
+ "length_s": 0.157,
+ "typ_z": 3.25,
+ "nn_cos": 0.792,
+ "tone_db": 9.19,
+ "tone_ratio": 0.86,
+ "n_outside": 1,
+ "outside": [
+ [
+ "39 Hz",
+ -24.8,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 65.0,
+ "active_s": 0.099,
+ "onset_s": 0.006,
+ "env_oct": 0.437,
+ "curve_db": 14.99,
+ "t40_ratio": 1.529,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.623,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 42.5,
+ "active_s": 0.061
+ },
+ "phrase_z": 2.93,
+ "phrase_min": 2.79,
+ "margin": 0.83,
+ "wins": 1.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Coin (dropping)",
+ 0.274
+ ],
+ [
+ "Chink, clink",
+ 0.077
+ ],
+ [
+ "Speech",
+ 0.069
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.0,
+ "floor_db": -49.6,
+ "stereo_corr": 1.0,
+ "darken": -0.07,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -105.1,
+ "first_sample": 0.0,
+ "tail_db": -60.2,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/bench/fs-431419_2.wav",
+ "length_s": 0.054,
+ "typ_z": 2.85,
+ "nn_cos": 0.792,
+ "tone_db": 9.85,
+ "tone_ratio": 0.95,
+ "n_outside": 3,
+ "outside": [
+ [
+ "10.1 kHz",
+ 15.6,
+ "above"
+ ],
+ [
+ "12.7 kHz",
+ 15.1,
+ "above"
+ ],
+ [
+ "2 kHz",
+ -14.9,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 40.0,
+ "active_s": 0.04,
+ "onset_s": 0.005,
+ "env_oct": 0.459,
+ "curve_db": 12.03,
+ "t40_ratio": 0.727,
+ "attack_ratio": 1.0,
+ "active_ratio": 0.53,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 55.0,
+ "active_s": 0.075
+ },
+ "phrase_z": 3.05,
+ "phrase_min": 2.77,
+ "margin": 0.27,
+ "wins": 0.5,
+ "best_confusion": "a finger snap",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.064
+ ],
+ [
+ "Music",
+ 0.041
+ ],
+ [
+ "Speech",
+ 0.031
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -40.8,
+ "stereo_corr": 1.0,
+ "darken": 0.42,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -89.0,
+ "first_sample": 0.0,
+ "tail_db": -42.6,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_rev": [
+ {
+ "file": "E:/earshot/out/refs/click/bench/mk-1124_1.wav",
+ "length_s": 0.055,
+ "typ_z": -0.27,
+ "nn_cos": 0.755,
+ "tone_db": 16.45,
+ "tone_ratio": 1.77,
+ "n_outside": 1,
+ "outside": [
+ [
+ "62 Hz",
+ -36.9,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 25.0,
+ "active_s": 0.022,
+ "onset_s": 0.008,
+ "env_oct": 0.946,
+ "curve_db": 10.93,
+ "t40_ratio": 0.417,
+ "attack_ratio": 1.0,
+ "active_ratio": 0.336,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 60.0,
+ "active_s": 0.066
+ },
+ "phrase_z": 2.67,
+ "phrase_min": 2.65,
+ "margin": 0.63,
+ "wins": 1.0,
+ "best_confusion": "a drop of water",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.063
+ ],
+ [
+ "Music",
+ 0.045
+ ],
+ [
+ "Speech",
+ 0.032
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -68.2,
+ "stereo_corr": 1.0,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -94.4,
+ "first_sample": 0.0,
+ "tail_db": -71.2,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/bench/mk-2364_1.wav",
+ "length_s": 0.21,
+ "typ_z": -3.09,
+ "nn_cos": 0.604,
+ "tone_db": 7.94,
+ "tone_ratio": 0.85,
+ "n_outside": 3,
+ "outside": [
+ [
+ "630 Hz",
+ 14.3,
+ "above"
+ ],
+ [
+ "794 Hz",
+ 10.9,
+ "above"
+ ],
+ [
+ "500 Hz",
+ 10.7,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.08,
+ "onset_s": 0.005,
+ "env_oct": 0.235,
+ "curve_db": 17.87,
+ "t40_ratio": 0.75,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.221,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 60.0,
+ "active_s": 0.066
+ },
+ "phrase_z": 2.4,
+ "phrase_min": 2.21,
+ "margin": -0.06,
+ "wins": 0.5,
+ "best_confusion": "a drop of water",
+ "as_top": [
+ [
+ "Plop",
+ 0.667
+ ],
+ [
+ "Music",
+ 0.045
+ ],
+ [
+ "Sound effect",
+ 0.042
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.001,
+ "floor_db": -47.3,
+ "stereo_corr": 1.0,
+ "darken": -2.12,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -53.9,
+ "first_sample": 0.0,
+ "tail_db": -59.7,
+ "lead_s": 0.004,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/bench/fs-610524_1.wav",
+ "length_s": 0.083,
+ "typ_z": -0.82,
+ "nn_cos": 0.703,
+ "tone_db": 8.92,
+ "tone_ratio": 0.96,
+ "n_outside": 2,
+ "outside": [
+ [
+ "16 kHz",
+ 26.7,
+ "above"
+ ],
+ [
+ "198 Hz",
+ 16.3,
+ "above"
+ ]
+ ],
+ "attack_ms": 16.0,
+ "t40_ms": 35.0,
+ "active_s": 0.051,
+ "onset_s": 0.014,
+ "env_oct": 1.713,
+ "curve_db": 13.65,
+ "t40_ratio": 0.583,
+ "attack_ratio": 16.0,
+ "active_ratio": 0.779,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 60.0,
+ "active_s": 0.066
+ },
+ "phrase_z": 2.92,
+ "phrase_min": 2.89,
+ "margin": 1.38,
+ "wins": 1.0,
+ "best_confusion": "a finger snap",
+ "as_top": [
+ [
+ "Speech",
+ 0.053
+ ],
+ [
+ "Music",
+ 0.053
+ ],
+ [
+ "Sound effect",
+ 0.04
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -44.8,
+ "stereo_corr": -0.129,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -61.6,
+ "first_sample": 0.0,
+ "tail_db": -53.6,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/bench/fs-461799_1.wav",
+ "length_s": 0.305,
+ "typ_z": -1.52,
+ "nn_cos": 0.708,
+ "tone_db": 7.89,
+ "tone_ratio": 0.85,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 149.0,
+ "t40_ms": 110.0,
+ "active_s": 0.255,
+ "onset_s": 0.03,
+ "env_oct": 3.352,
+ "curve_db": 25.04,
+ "t40_ratio": 1.833,
+ "attack_ratio": 149.0,
+ "active_ratio": 3.893,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 60.0,
+ "active_s": 0.066
+ },
+ "phrase_z": 2.34,
+ "phrase_min": 2.33,
+ "margin": -0.44,
+ "wins": 0.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.578
+ ],
+ [
+ "Clapping",
+ 0.447
+ ],
+ [
+ "Speech",
+ 0.045
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.024,
+ "floor_db": -38.9,
+ "stereo_corr": 0.824,
+ "darken": -0.35,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -74.6,
+ "first_sample": 0.0,
+ "tail_db": -48.8,
+ "lead_s": 0.005,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/bench/fs-461799_2.wav",
+ "length_s": 0.109,
+ "typ_z": -1.73,
+ "nn_cos": 0.713,
+ "tone_db": 9.67,
+ "tone_ratio": 1.04,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 1.0,
+ "t40_ms": 70.0,
+ "active_s": 0.071,
+ "onset_s": 0.019,
+ "env_oct": 0.113,
+ "curve_db": 11.56,
+ "t40_ratio": 1.167,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.084,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 60.0,
+ "active_s": 0.066
+ },
+ "phrase_z": 2.51,
+ "phrase_min": 2.48,
+ "margin": -0.19,
+ "wins": 0.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Speech",
+ 0.114
+ ],
+ [
+ "Music",
+ 0.034
+ ],
+ [
+ "Single-lens reflex camera",
+ 0.028
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.003,
+ "floor_db": -40.0,
+ "stereo_corr": 0.948,
+ "darken": 0.13,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -90.1,
+ "first_sample": 0.0,
+ "tail_db": -50.3,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/bench/fs-431419_1.wav",
+ "length_s": 0.157,
+ "typ_z": -0.02,
+ "nn_cos": 0.737,
+ "tone_db": 11.24,
+ "tone_ratio": 1.21,
+ "n_outside": 2,
+ "outside": [
+ [
+ "16 kHz",
+ 28.7,
+ "above"
+ ],
+ [
+ "39 Hz",
+ -27.2,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 65.0,
+ "active_s": 0.099,
+ "onset_s": 0.006,
+ "env_oct": 0.237,
+ "curve_db": 11.42,
+ "t40_ratio": 1.083,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.511,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 60.0,
+ "active_s": 0.066
+ },
+ "phrase_z": 2.93,
+ "phrase_min": 2.79,
+ "margin": 0.83,
+ "wins": 1.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Coin (dropping)",
+ 0.274
+ ],
+ [
+ "Chink, clink",
+ 0.077
+ ],
+ [
+ "Speech",
+ 0.069
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.0,
+ "floor_db": -49.6,
+ "stereo_corr": 1.0,
+ "darken": -0.07,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -105.1,
+ "first_sample": 0.0,
+ "tail_db": -60.2,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/bench/fs-431419_2.wav",
+ "length_s": 0.054,
+ "typ_z": 0.61,
+ "nn_cos": 0.807,
+ "tone_db": 11.37,
+ "tone_ratio": 1.22,
+ "n_outside": 4,
+ "outside": [
+ [
+ "62 Hz",
+ -36.9,
+ "below"
+ ],
+ [
+ "16 kHz",
+ 26.3,
+ "above"
+ ],
+ [
+ "12.7 kHz",
+ 12.6,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 40.0,
+ "active_s": 0.04,
+ "onset_s": 0.005,
+ "env_oct": 0.432,
+ "curve_db": 9.02,
+ "t40_ratio": 0.667,
+ "attack_ratio": 1.0,
+ "active_ratio": 0.611,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 60.0,
+ "active_s": 0.066
+ },
+ "phrase_z": 3.05,
+ "phrase_min": 2.77,
+ "margin": 0.27,
+ "wins": 0.5,
+ "best_confusion": "a finger snap",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.064
+ ],
+ [
+ "Music",
+ 0.041
+ ],
+ [
+ "Speech",
+ 0.031
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -40.8,
+ "stereo_corr": 1.0,
+ "darken": 0.42,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -89.0,
+ "first_sample": 0.0,
+ "tail_db": -42.6,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity_bench": {
+ "n": 7,
+ "cos_mean": 0.576,
+ "cos_max": 0.792,
+ "tone_min_db": 6.13,
+ "tone_mean_db": 12.82,
+ "near_duplicates": []
+ },
+ "diversity_forge": {
+ "n": 8,
+ "cos_mean": 0.651,
+ "cos_max": 0.92,
+ "tone_min_db": 4.52,
+ "tone_mean_db": 16.52,
+ "near_duplicates": []
+ },
+ "summary": {
+ "made": {},
+ "real_cross": {
+ "n": 8,
+ "typ_z": 0.16,
+ "tone_db": 11.33,
+ "env_oct": 0.22,
+ "curve_db": 9.92,
+ "phrase_z": 2.77,
+ "margin": 0.53,
+ "as_p": 0.0,
+ "tone_ratio": 1.15,
+ "t40_ratio": 1.33,
+ "wins": 0.75,
+ "floor_db": -42.85,
+ "stereo_corr": 0.99,
+ "darken": -0.12,
+ "as_names": 0.12,
+ "artifacts": 0,
+ "var_cos": 0.651
+ },
+ "real_loo": {
+ "n": 7,
+ "typ_z": -0.1,
+ "tone_db": 9.85,
+ "env_oct": 0.46,
+ "curve_db": 13.19,
+ "phrase_z": 2.67,
+ "margin": 0.27,
+ "as_p": 0.0,
+ "tone_ratio": 0.95,
+ "t40_ratio": 0.86,
+ "wins": 0.5,
+ "floor_db": -44.8,
+ "stereo_corr": 1.0,
+ "darken": -0.07,
+ "as_names": 0.0,
+ "artifacts": 0,
+ "var_cos": 0.576
+ }
+ }
+ },
+ "downlifter": {
+ "n_bench": 0,
+ "n_forge": 0,
+ "note": "empty: the licensed library has no downlifter, real or designed (no reversing or pitching of other sounds here: trim + fade only). Benchmark made downlifters on phrasings, AudioSet and artifacts only.",
+ "made": [],
+ "summary": {}
+ },
+ "glitch": {
+ "n_bench": 0,
+ "n_forge": 0,
+ "note": "empty: no glitch material, real or designed, in the licensed library. Benchmark made glitches on phrasings, AudioSet and artifacts only.",
+ "made": [],
+ "summary": {}
+ },
+ "hover_tick": {
+ "n_bench": 6,
+ "n_forge": 6,
+ "note": "",
+ "bench_refs": [
+ "E:/earshot/out/refs/hover_tick/bench/fs-401803_1.wav",
+ "E:/earshot/out/refs/hover_tick/bench/fs-401803_2.wav",
+ "E:/earshot/out/refs/hover_tick/bench/fs-401803_3.wav",
+ "E:/earshot/out/refs/hover_tick/bench/fs-873306_1.wav",
+ "E:/earshot/out/refs/hover_tick/bench/fs-873306_2.wav",
+ "E:/earshot/out/refs/hover_tick/bench/fs-326470_1.wav"
+ ],
+ "made": [],
+ "diversity": {
+ "n": 0
+ },
+ "real_cross": [
+ {
+ "file": "E:/earshot/out/refs/hover_tick/forge/kit-ui_tick_01.wav",
+ "length_s": 0.052,
+ "typ_z": 2.39,
+ "nn_cos": 0.785,
+ "tone_db": 19.65,
+ "tone_ratio": 1.27,
+ "n_outside": 9,
+ "outside": [
+ [
+ "1 kHz",
+ -25.9,
+ "below"
+ ],
+ [
+ "794 Hz",
+ -25.8,
+ "below"
+ ],
+ [
+ "1.26 kHz",
+ -23.7,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 30.0,
+ "active_s": 0.038,
+ "onset_s": 0.007,
+ "env_oct": 1.656,
+ "curve_db": 8.94,
+ "t40_ratio": 0.6,
+ "attack_ratio": 0.087,
+ "active_ratio": 0.613,
+ "ref_median": {
+ "attack_ms": 11.5,
+ "t40_ms": 50.0,
+ "active_s": 0.062
+ },
+ "phrase_z": 2.68,
+ "phrase_min": 1.92,
+ "margin": 1.58,
+ "wins": 0.5,
+ "best_confusion": "a clock ticking",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.061
+ ],
+ [
+ "Music",
+ 0.043
+ ],
+ [
+ "Speech",
+ 0.034
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -45.2,
+ "stereo_corr": 1.0,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -112.4,
+ "first_sample": 0.0,
+ "tail_db": -42.3,
+ "lead_s": 0.006,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/hover_tick/forge/kit-ui_tick_02.wav",
+ "length_s": 0.097,
+ "typ_z": 0.08,
+ "nn_cos": 0.743,
+ "tone_db": 20.57,
+ "tone_ratio": 1.33,
+ "n_outside": 7,
+ "outside": [
+ [
+ "250 Hz",
+ -34.9,
+ "below"
+ ],
+ [
+ "397 Hz",
+ -23.6,
+ "below"
+ ],
+ [
+ "500 Hz",
+ -20.5,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 90.0,
+ "active_s": 0.087,
+ "onset_s": 0.008,
+ "env_oct": 1.62,
+ "curve_db": 9.81,
+ "t40_ratio": 1.8,
+ "attack_ratio": 0.087,
+ "active_ratio": 1.403,
+ "ref_median": {
+ "attack_ms": 11.5,
+ "t40_ms": 50.0,
+ "active_s": 0.062
+ },
+ "phrase_z": 2.49,
+ "phrase_min": 1.56,
+ "margin": 0.48,
+ "wins": 0.5,
+ "best_confusion": "a clock ticking",
+ "as_top": [
+ [
+ "Coin (dropping)",
+ 0.118
+ ],
+ [
+ "Speech",
+ 0.078
+ ],
+ [
+ "Chink, clink",
+ 0.054
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.0,
+ "floor_db": -37.9,
+ "stereo_corr": 1.0,
+ "darken": -0.14,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -104.7,
+ "first_sample": 0.0,
+ "tail_db": -39.0,
+ "lead_s": 0.005,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/hover_tick/forge/fs-376195_1.wav",
+ "length_s": 0.036,
+ "typ_z": 0.77,
+ "nn_cos": 0.749,
+ "tone_db": 21.82,
+ "tone_ratio": 1.41,
+ "n_outside": 12,
+ "outside": [
+ [
+ "250 Hz",
+ -34.9,
+ "below"
+ ],
+ [
+ "315 Hz",
+ -32.7,
+ "below"
+ ],
+ [
+ "1 kHz",
+ -26.1,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 25.0,
+ "active_s": 0.024,
+ "onset_s": 0.006,
+ "env_oct": 1.964,
+ "curve_db": 12.55,
+ "t40_ratio": 0.5,
+ "attack_ratio": 0.087,
+ "active_ratio": 0.387,
+ "ref_median": {
+ "attack_ms": 11.5,
+ "t40_ms": 50.0,
+ "active_s": 0.062
+ },
+ "phrase_z": 2.53,
+ "phrase_min": 1.87,
+ "margin": 1.11,
+ "wins": 0.5,
+ "best_confusion": "a cricket chirp",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.065
+ ],
+ [
+ "Music",
+ 0.043
+ ],
+ [
+ "Speech",
+ 0.029
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -47.7,
+ "stereo_corr": 1.0,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -88.9,
+ "first_sample": 0.0,
+ "tail_db": -42.0,
+ "lead_s": 0.004,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/hover_tick/forge/fs-376195_2.wav",
+ "length_s": 0.035,
+ "typ_z": 0.1,
+ "nn_cos": 0.769,
+ "tone_db": 23.48,
+ "tone_ratio": 1.52,
+ "n_outside": 13,
+ "outside": [
+ [
+ "250 Hz",
+ -34.9,
+ "below"
+ ],
+ [
+ "315 Hz",
+ -34.1,
+ "below"
+ ],
+ [
+ "397 Hz",
+ -31.5,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 25.0,
+ "active_s": 0.024,
+ "onset_s": 0.006,
+ "env_oct": 1.964,
+ "curve_db": 12.74,
+ "t40_ratio": 0.5,
+ "attack_ratio": 0.087,
+ "active_ratio": 0.387,
+ "ref_median": {
+ "attack_ms": 11.5,
+ "t40_ms": 50.0,
+ "active_s": 0.062
+ },
+ "phrase_z": 2.68,
+ "phrase_min": 2.02,
+ "margin": 1.21,
+ "wins": 0.5,
+ "best_confusion": "a clock ticking",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.064
+ ],
+ [
+ "Music",
+ 0.041
+ ],
+ [
+ "Speech",
+ 0.028
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -41.9,
+ "stereo_corr": 1.0,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -102.4,
+ "first_sample": 0.0,
+ "tail_db": -40.2,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/hover_tick/forge/fs-329098_1.wav",
+ "length_s": 0.107,
+ "typ_z": -2.08,
+ "nn_cos": 0.674,
+ "tone_db": 19.29,
+ "tone_ratio": 1.25,
+ "n_outside": 4,
+ "outside": [
+ [
+ "16 kHz",
+ -17.5,
+ "below"
+ ],
+ [
+ "794 Hz",
+ -15.2,
+ "below"
+ ],
+ [
+ "500 Hz",
+ -14.1,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 85.0,
+ "active_s": 0.085,
+ "onset_s": 0.005,
+ "env_oct": 1.581,
+ "curve_db": 9.95,
+ "t40_ratio": 1.7,
+ "attack_ratio": 0.087,
+ "active_ratio": 1.371,
+ "ref_median": {
+ "attack_ms": 11.5,
+ "t40_ms": 50.0,
+ "active_s": 0.062
+ },
+ "phrase_z": 2.44,
+ "phrase_min": 1.48,
+ "margin": 0.68,
+ "wins": 0.5,
+ "best_confusion": "a clock ticking",
+ "as_top": [
+ [
+ "Dishes, pots, and pans",
+ 0.086
+ ],
+ [
+ "Cutlery, silverware",
+ 0.071
+ ],
+ [
+ "Chink, clink",
+ 0.063
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.0,
+ "floor_db": -46.3,
+ "stereo_corr": 1.0,
+ "darken": -0.09,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -91.8,
+ "first_sample": 0.0,
+ "tail_db": -52.5,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/hover_tick/forge/fs-329098_2.wav",
+ "length_s": 0.079,
+ "typ_z": -1.73,
+ "nn_cos": 0.686,
+ "tone_db": 22.7,
+ "tone_ratio": 1.47,
+ "n_outside": 8,
+ "outside": [
+ [
+ "250 Hz",
+ -34.9,
+ "below"
+ ],
+ [
+ "315 Hz",
+ -34.1,
+ "below"
+ ],
+ [
+ "397 Hz",
+ -31.5,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 60.0,
+ "active_s": 0.06,
+ "onset_s": 0.005,
+ "env_oct": 1.278,
+ "curve_db": 6.66,
+ "t40_ratio": 1.2,
+ "attack_ratio": 0.087,
+ "active_ratio": 0.968,
+ "ref_median": {
+ "attack_ms": 11.5,
+ "t40_ms": 50.0,
+ "active_s": 0.062
+ },
+ "phrase_z": 2.38,
+ "phrase_min": 1.44,
+ "margin": 0.53,
+ "wins": 0.5,
+ "best_confusion": "a clock ticking",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.09
+ ],
+ [
+ "Coin (dropping)",
+ 0.068
+ ],
+ [
+ "Speech",
+ 0.047
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -40.4,
+ "stereo_corr": 1.0,
+ "darken": 0.0,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -110.6,
+ "first_sample": 0.0,
+ "tail_db": -48.4,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_loo": [
+ {
+ "file": "E:/earshot/out/refs/hover_tick/bench/fs-401803_1.wav",
+ "length_s": 0.062,
+ "typ_z": 2.8,
+ "nn_cos": 0.739,
+ "tone_db": 14.68,
+ "tone_ratio": 0.9,
+ "n_outside": 1,
+ "outside": [
+ [
+ "794 Hz",
+ 8.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 29.0,
+ "t40_ms": 30.0,
+ "active_s": 0.057,
+ "onset_s": 0.003,
+ "env_oct": 1.207,
+ "curve_db": 25.27,
+ "t40_ratio": 0.462,
+ "attack_ratio": 4.833,
+ "active_ratio": 0.851,
+ "ref_median": {
+ "attack_ms": 6.0,
+ "t40_ms": 65.0,
+ "active_s": 0.067
+ },
+ "phrase_z": 2.01,
+ "phrase_min": 1.52,
+ "margin": 0.76,
+ "wins": 0.0,
+ "best_confusion": "a typewriter",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.051
+ ],
+ [
+ "Speech",
+ 0.034
+ ],
+ [
+ "Music",
+ 0.031
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -20.3,
+ "stereo_corr": 0.097,
+ "artifacts": [
+ "tail cut"
+ ],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -31.5,
+ "first_sample": 0.0,
+ "tail_db": -23.7,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/hover_tick/bench/fs-401803_2.wav",
+ "length_s": 0.044,
+ "typ_z": 0.12,
+ "nn_cos": 0.739,
+ "tone_db": 14.36,
+ "tone_ratio": 0.88,
+ "n_outside": 6,
+ "outside": [
+ [
+ "198 Hz",
+ 21.6,
+ "above"
+ ],
+ [
+ "250 Hz",
+ 19.3,
+ "above"
+ ],
+ [
+ "397 Hz",
+ 18.2,
+ "above"
+ ]
+ ],
+ "attack_ms": 6.0,
+ "t40_ms": 35.0,
+ "active_s": 0.037,
+ "onset_s": 0.003,
+ "env_oct": 1.084,
+ "curve_db": 16.66,
+ "t40_ratio": 0.538,
+ "attack_ratio": 0.353,
+ "active_ratio": 0.552,
+ "ref_median": {
+ "attack_ms": 17.0,
+ "t40_ms": 65.0,
+ "active_s": 0.067
+ },
+ "phrase_z": 1.57,
+ "phrase_min": 0.92,
+ "margin": 0.23,
+ "wins": 0.0,
+ "best_confusion": "a drop of water",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.063
+ ],
+ [
+ "Music",
+ 0.041
+ ],
+ [
+ "Speech",
+ 0.027
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -22.2,
+ "stereo_corr": 0.491,
+ "artifacts": [
+ "dc offset",
+ "tail cut"
+ ],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -29.2,
+ "first_sample": 0.0,
+ "tail_db": -27.7,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/hover_tick/bench/fs-401803_3.wav",
+ "length_s": 0.121,
+ "typ_z": 0.76,
+ "nn_cos": 0.715,
+ "tone_db": 16.24,
+ "tone_ratio": 1.11,
+ "n_outside": 3,
+ "outside": [
+ [
+ "79 Hz",
+ 52.9,
+ "above"
+ ],
+ [
+ "39 Hz",
+ 36.1,
+ "above"
+ ],
+ [
+ "62 Hz",
+ 22.3,
+ "above"
+ ]
+ ],
+ "attack_ms": 17.0,
+ "t40_ms": 65.0,
+ "active_s": 0.11,
+ "onset_s": 0.005,
+ "env_oct": 1.115,
+ "curve_db": 28.35,
+ "t40_ratio": 1.857,
+ "attack_ratio": 2.833,
+ "active_ratio": 1.93,
+ "ref_median": {
+ "attack_ms": 6.0,
+ "t40_ms": 35.0,
+ "active_s": 0.057
+ },
+ "phrase_z": 0.8,
+ "phrase_min": -0.02,
+ "margin": -0.72,
+ "wins": 0.0,
+ "best_confusion": "a typewriter",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.053
+ ],
+ [
+ "Music",
+ 0.053
+ ],
+ [
+ "Speech",
+ 0.034
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.001,
+ "floor_db": -36.2,
+ "stereo_corr": 0.552,
+ "darken": -0.79,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -44.3,
+ "first_sample": 0.0,
+ "tail_db": -41.7,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/hover_tick/bench/fs-873306_1.wav",
+ "length_s": 0.03,
+ "typ_z": 0.9,
+ "nn_cos": 0.847,
+ "tone_db": 24.22,
+ "tone_ratio": 1.65,
+ "n_outside": 12,
+ "outside": [
+ [
+ "157 Hz",
+ -47.1,
+ "below"
+ ],
+ [
+ "198 Hz",
+ -46.7,
+ "below"
+ ],
+ [
+ "125 Hz",
+ -45.6,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 25.0,
+ "active_s": 0.026,
+ "onset_s": 0.004,
+ "env_oct": 2.277,
+ "curve_db": 18.72,
+ "t40_ratio": 0.385,
+ "attack_ratio": 0.059,
+ "active_ratio": 0.388,
+ "ref_median": {
+ "attack_ms": 17.0,
+ "t40_ms": 65.0,
+ "active_s": 0.067
+ },
+ "phrase_z": 2.82,
+ "phrase_min": 2.29,
+ "margin": 1.42,
+ "wins": 0.5,
+ "best_confusion": "a clock ticking",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.058
+ ],
+ [
+ "Music",
+ 0.038
+ ],
+ [
+ "Speech",
+ 0.03
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -21.0,
+ "stereo_corr": 0.713,
+ "artifacts": [
+ "tail cut"
+ ],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -102.4,
+ "first_sample": 0.0,
+ "tail_db": -19.2,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/hover_tick/bench/fs-873306_2.wav",
+ "length_s": 0.081,
+ "typ_z": 0.17,
+ "nn_cos": 0.847,
+ "tone_db": 19.73,
+ "tone_ratio": 1.34,
+ "n_outside": 3,
+ "outside": [
+ [
+ "4 kHz",
+ 13.9,
+ "above"
+ ],
+ [
+ "3.17 kHz",
+ 9.0,
+ "above"
+ ],
+ [
+ "1 kHz",
+ 4.8,
+ "above"
+ ]
+ ],
+ "attack_ms": 19.0,
+ "t40_ms": 75.0,
+ "active_s": 0.076,
+ "onset_s": 0.004,
+ "env_oct": 1.059,
+ "curve_db": 20.07,
+ "t40_ratio": 2.143,
+ "attack_ratio": 3.167,
+ "active_ratio": 1.333,
+ "ref_median": {
+ "attack_ms": 6.0,
+ "t40_ms": 35.0,
+ "active_s": 0.057
+ },
+ "phrase_z": 2.57,
+ "phrase_min": 2.02,
+ "margin": 1.4,
+ "wins": 0.5,
+ "best_confusion": "a typewriter",
+ "as_top": [
+ [
+ "Speech",
+ 0.085
+ ],
+ [
+ "Single-lens reflex camera",
+ 0.08
+ ],
+ [
+ "Coin (dropping)",
+ 0.078
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.003,
+ "floor_db": -27.6,
+ "stereo_corr": 0.169,
+ "darken": -0.01,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -94.9,
+ "first_sample": 0.0,
+ "tail_db": -31.8,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/hover_tick/bench/fs-326470_1.wav",
+ "length_s": 0.093,
+ "typ_z": -0.81,
+ "nn_cos": 0.688,
+ "tone_db": 14.37,
+ "tone_ratio": 0.88,
+ "n_outside": 1,
+ "outside": [
+ [
+ "31 Hz",
+ 37.6,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 70.0,
+ "active_s": 0.067,
+ "onset_s": 0.018,
+ "env_oct": 1.774,
+ "curve_db": 16.62,
+ "t40_ratio": 2.0,
+ "attack_ratio": 0.059,
+ "active_ratio": 1.175,
+ "ref_median": {
+ "attack_ms": 17.0,
+ "t40_ms": 35.0,
+ "active_s": 0.057
+ },
+ "phrase_z": 2.0,
+ "phrase_min": 1.43,
+ "margin": 0.4,
+ "wins": 0.5,
+ "best_confusion": "a typewriter",
+ "as_top": [
+ [
+ "Speech",
+ 0.128
+ ],
+ [
+ "Sound effect",
+ 0.036
+ ],
+ [
+ "Finger snapping",
+ 0.035
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.001,
+ "floor_db": -34.1,
+ "stereo_corr": 1.0,
+ "darken": 0.08,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -55.9,
+ "first_sample": 0.0,
+ "tail_db": -39.1,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_rev": [
+ {
+ "file": "E:/earshot/out/refs/hover_tick/bench/fs-401803_1.wav",
+ "length_s": 0.062,
+ "typ_z": -11.43,
+ "nn_cos": 0.697,
+ "tone_db": 27.79,
+ "tone_ratio": 2.67,
+ "n_outside": 17,
+ "outside": [
+ [
+ "198 Hz",
+ 47.6,
+ "above"
+ ],
+ [
+ "99 Hz",
+ 46.9,
+ "above"
+ ],
+ [
+ "157 Hz",
+ 46.8,
+ "above"
+ ]
+ ],
+ "attack_ms": 29.0,
+ "t40_ms": 30.0,
+ "active_s": 0.057,
+ "onset_s": 0.003,
+ "env_oct": 1.887,
+ "curve_db": 21.45,
+ "t40_ratio": 0.667,
+ "attack_ratio": 29.0,
+ "active_ratio": 1.163,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.049
+ },
+ "phrase_z": 2.01,
+ "phrase_min": 1.52,
+ "margin": 0.76,
+ "wins": 0.0,
+ "best_confusion": "a typewriter",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.051
+ ],
+ [
+ "Speech",
+ 0.034
+ ],
+ [
+ "Music",
+ 0.031
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -20.3,
+ "stereo_corr": 0.097,
+ "artifacts": [
+ "tail cut"
+ ],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -31.5,
+ "first_sample": 0.0,
+ "tail_db": -23.7,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/hover_tick/bench/fs-401803_2.wav",
+ "length_s": 0.044,
+ "typ_z": -18.12,
+ "nn_cos": 0.531,
+ "tone_db": 28.45,
+ "tone_ratio": 2.73,
+ "n_outside": 15,
+ "outside": [
+ [
+ "198 Hz",
+ 51.6,
+ "above"
+ ],
+ [
+ "250 Hz",
+ 49.6,
+ "above"
+ ],
+ [
+ "157 Hz",
+ 49.3,
+ "above"
+ ]
+ ],
+ "attack_ms": 6.0,
+ "t40_ms": 35.0,
+ "active_s": 0.037,
+ "onset_s": 0.003,
+ "env_oct": 1.118,
+ "curve_db": 8.21,
+ "t40_ratio": 0.778,
+ "attack_ratio": 6.0,
+ "active_ratio": 0.755,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.049
+ },
+ "phrase_z": 1.57,
+ "phrase_min": 0.92,
+ "margin": 0.23,
+ "wins": 0.0,
+ "best_confusion": "a drop of water",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.063
+ ],
+ [
+ "Music",
+ 0.041
+ ],
+ [
+ "Speech",
+ 0.027
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -22.2,
+ "stereo_corr": 0.491,
+ "artifacts": [
+ "dc offset",
+ "tail cut"
+ ],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -29.2,
+ "first_sample": 0.0,
+ "tail_db": -27.7,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/hover_tick/bench/fs-401803_3.wav",
+ "length_s": 0.121,
+ "typ_z": -11.36,
+ "nn_cos": 0.669,
+ "tone_db": 28.43,
+ "tone_ratio": 2.73,
+ "n_outside": 15,
+ "outside": [
+ [
+ "79 Hz",
+ 49.5,
+ "above"
+ ],
+ [
+ "157 Hz",
+ 48.8,
+ "above"
+ ],
+ [
+ "62 Hz",
+ 47.7,
+ "above"
+ ]
+ ],
+ "attack_ms": 17.0,
+ "t40_ms": 65.0,
+ "active_s": 0.11,
+ "onset_s": 0.005,
+ "env_oct": 1.928,
+ "curve_db": 26.02,
+ "t40_ratio": 1.444,
+ "attack_ratio": 17.0,
+ "active_ratio": 2.245,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.049
+ },
+ "phrase_z": 0.8,
+ "phrase_min": -0.02,
+ "margin": -0.72,
+ "wins": 0.0,
+ "best_confusion": "a typewriter",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.053
+ ],
+ [
+ "Music",
+ 0.053
+ ],
+ [
+ "Speech",
+ 0.034
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.001,
+ "floor_db": -36.2,
+ "stereo_corr": 0.552,
+ "darken": -0.79,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -44.3,
+ "first_sample": 0.0,
+ "tail_db": -41.7,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/hover_tick/bench/fs-873306_1.wav",
+ "length_s": 0.03,
+ "typ_z": -5.16,
+ "nn_cos": 0.769,
+ "tone_db": 7.88,
+ "tone_ratio": 0.76,
+ "n_outside": 4,
+ "outside": [
+ [
+ "794 Hz",
+ 21.0,
+ "above"
+ ],
+ [
+ "3.17 kHz",
+ -14.3,
+ "below"
+ ],
+ [
+ "5.04 kHz",
+ -10.9,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 25.0,
+ "active_s": 0.026,
+ "onset_s": 0.004,
+ "env_oct": 0.587,
+ "curve_db": 9.87,
+ "t40_ratio": 0.556,
+ "attack_ratio": 1.0,
+ "active_ratio": 0.531,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.049
+ },
+ "phrase_z": 2.82,
+ "phrase_min": 2.29,
+ "margin": 1.42,
+ "wins": 0.5,
+ "best_confusion": "a clock ticking",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.058
+ ],
+ [
+ "Music",
+ 0.038
+ ],
+ [
+ "Speech",
+ 0.03
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -21.0,
+ "stereo_corr": 0.713,
+ "artifacts": [
+ "tail cut"
+ ],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -102.4,
+ "first_sample": 0.0,
+ "tail_db": -19.2,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/hover_tick/bench/fs-873306_2.wav",
+ "length_s": 0.081,
+ "typ_z": -10.29,
+ "nn_cos": 0.685,
+ "tone_db": 13.77,
+ "tone_ratio": 1.32,
+ "n_outside": 12,
+ "outside": [
+ [
+ "315 Hz",
+ 29.5,
+ "above"
+ ],
+ [
+ "250 Hz",
+ 26.4,
+ "above"
+ ],
+ [
+ "1 kHz",
+ 24.2,
+ "above"
+ ]
+ ],
+ "attack_ms": 19.0,
+ "t40_ms": 75.0,
+ "active_s": 0.076,
+ "onset_s": 0.004,
+ "env_oct": 1.873,
+ "curve_db": 14.77,
+ "t40_ratio": 1.667,
+ "attack_ratio": 19.0,
+ "active_ratio": 1.551,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.049
+ },
+ "phrase_z": 2.57,
+ "phrase_min": 2.02,
+ "margin": 1.4,
+ "wins": 0.5,
+ "best_confusion": "a typewriter",
+ "as_top": [
+ [
+ "Speech",
+ 0.085
+ ],
+ [
+ "Single-lens reflex camera",
+ 0.08
+ ],
+ [
+ "Coin (dropping)",
+ 0.078
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.003,
+ "floor_db": -27.6,
+ "stereo_corr": 0.169,
+ "darken": -0.01,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -94.9,
+ "first_sample": 0.0,
+ "tail_db": -31.8,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/hover_tick/bench/fs-326470_1.wav",
+ "length_s": 0.093,
+ "typ_z": -7.19,
+ "nn_cos": 0.785,
+ "tone_db": 21.28,
+ "tone_ratio": 2.04,
+ "n_outside": 16,
+ "outside": [
+ [
+ "31 Hz",
+ 37.6,
+ "above"
+ ],
+ [
+ "39 Hz",
+ 33.1,
+ "above"
+ ],
+ [
+ "50 Hz",
+ 32.0,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 70.0,
+ "active_s": 0.067,
+ "onset_s": 0.018,
+ "env_oct": 0.363,
+ "curve_db": 12.79,
+ "t40_ratio": 1.556,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.367,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.049
+ },
+ "phrase_z": 2.0,
+ "phrase_min": 1.43,
+ "margin": 0.4,
+ "wins": 0.5,
+ "best_confusion": "a typewriter",
+ "as_top": [
+ [
+ "Speech",
+ 0.128
+ ],
+ [
+ "Sound effect",
+ 0.036
+ ],
+ [
+ "Finger snapping",
+ 0.035
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.001,
+ "floor_db": -34.1,
+ "stereo_corr": 1.0,
+ "darken": 0.08,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -55.9,
+ "first_sample": 0.0,
+ "tail_db": -39.1,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity_bench": {
+ "n": 6,
+ "cos_mean": 0.626,
+ "cos_max": 0.847,
+ "tone_min_db": 9.21,
+ "tone_mean_db": 18.42,
+ "near_duplicates": []
+ },
+ "diversity_forge": {
+ "n": 6,
+ "cos_mean": 0.799,
+ "cos_max": 0.952,
+ "tone_min_db": 3.57,
+ "tone_mean_db": 9.45,
+ "near_duplicates": []
+ },
+ "summary": {
+ "made": {},
+ "real_cross": {
+ "n": 6,
+ "typ_z": 0.09,
+ "tone_db": 21.2,
+ "env_oct": 1.64,
+ "curve_db": 9.88,
+ "phrase_z": 2.51,
+ "margin": 0.9,
+ "as_p": 0.0,
+ "tone_ratio": 1.37,
+ "t40_ratio": 0.9,
+ "wins": 0.5,
+ "floor_db": -43.55,
+ "stereo_corr": 1.0,
+ "darken": -0.09,
+ "as_names": 0.0,
+ "artifacts": 0,
+ "var_cos": 0.799
+ },
+ "real_loo": {
+ "n": 6,
+ "typ_z": 0.47,
+ "tone_db": 15.46,
+ "env_oct": 1.16,
+ "curve_db": 19.39,
+ "phrase_z": 2.0,
+ "margin": 0.58,
+ "as_p": 0.0,
+ "tone_ratio": 1.01,
+ "t40_ratio": 1.2,
+ "wins": 0.25,
+ "floor_db": -24.9,
+ "stereo_corr": 0.52,
+ "darken": -0.01,
+ "as_names": 0.0,
+ "artifacts": 4,
+ "var_cos": 0.626
+ }
+ }
+ },
+ "impact": {
+ "n_bench": 1,
+ "n_forge": 2,
+ "note": "thin: the library has three cinematic hits (all three were the v1 impact forge refs); one is held out as bench, so v1 impact bench numbers are optimistic.",
+ "bench_refs": [
+ "E:/earshot/out/refs/impact/bench/kit-impact_03.wav"
+ ],
+ "made": [],
+ "diversity": {
+ "n": 0
+ },
+ "real_cross": [
+ {
+ "file": "E:/earshot/out/refs/impact/forge/kit-impact_01.wav",
+ "length_s": 4.455,
+ "nn_cos": 0.449,
+ "tone_db": 7.04,
+ "tone_ratio": null,
+ "n_outside": 18,
+ "outside": [
+ [
+ "12.7 kHz",
+ -18.6,
+ "below"
+ ],
+ [
+ "16 kHz",
+ -17.5,
+ "below"
+ ],
+ [
+ "10.1 kHz",
+ -8.5,
+ "below"
+ ]
+ ],
+ "attack_ms": 10.0,
+ "t40_ms": 1310.0,
+ "active_s": 1.812,
+ "onset_s": 0.013,
+ "env_oct": 3.742,
+ "curve_db": 41.06,
+ "t40_ratio": 0.38,
+ "attack_ratio": 0.004,
+ "active_ratio": 0.276,
+ "ref_median": {
+ "attack_ms": 2505.0,
+ "t40_ms": 3450.0,
+ "active_s": 6.573
+ },
+ "phrase_z": 3.04,
+ "phrase_min": 2.25,
+ "margin": 0.69,
+ "wins": 0.5,
+ "best_confusion": "a bass drum",
+ "as_top": [
+ [
+ "Bang",
+ 0.479
+ ],
+ [
+ "Explosion",
+ 0.221
+ ],
+ [
+ "Burst, pop",
+ 0.206
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.479,
+ "floor_db": -88.8,
+ "stereo_corr": 0.338,
+ "darken": -0.56,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -69.1,
+ "first_sample": 0.0,
+ "tail_db": -130.9,
+ "lead_s": 0.007,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/impact/forge/kit-impact_02.wav",
+ "length_s": 6.5,
+ "nn_cos": 0.842,
+ "tone_db": 5.33,
+ "tone_ratio": null,
+ "n_outside": 15,
+ "outside": [
+ [
+ "50 Hz",
+ -13.3,
+ "below"
+ ],
+ [
+ "39 Hz",
+ -10.9,
+ "below"
+ ],
+ [
+ "31 Hz",
+ -8.6,
+ "below"
+ ]
+ ],
+ "attack_ms": 487.0,
+ "t40_ms": 1865.0,
+ "active_s": 4.565,
+ "onset_s": 0.505,
+ "env_oct": 1.259,
+ "curve_db": 22.9,
+ "t40_ratio": 0.541,
+ "attack_ratio": 0.194,
+ "active_ratio": 0.695,
+ "ref_median": {
+ "attack_ms": 2505.0,
+ "t40_ms": 3450.0,
+ "active_s": 6.573
+ },
+ "phrase_z": 3.03,
+ "phrase_min": 1.8,
+ "margin": 1.5,
+ "wins": 0.5,
+ "best_confusion": "a bass drum",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.808
+ ],
+ [
+ "Sound effect",
+ 0.204
+ ],
+ [
+ "Music",
+ 0.062
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.028,
+ "floor_db": -52.3,
+ "stereo_corr": 0.745,
+ "darken": -2.04,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -41.5,
+ "first_sample": 0.0,
+ "tail_db": -100.5,
+ "lead_s": 0.041,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_loo": [],
+ "real_rev": [
+ {
+ "file": "E:/earshot/out/refs/impact/bench/kit-impact_03.wav",
+ "length_s": 7.92,
+ "nn_cos": 0.842,
+ "tone_db": 5.13,
+ "tone_ratio": null,
+ "n_outside": 8,
+ "outside": [
+ [
+ "50 Hz",
+ 10.3,
+ "above"
+ ],
+ [
+ "4 kHz",
+ 7.3,
+ "above"
+ ],
+ [
+ "157 Hz",
+ -7.0,
+ "below"
+ ]
+ ],
+ "attack_ms": 2505.0,
+ "t40_ms": 3450.0,
+ "active_s": 6.573,
+ "onset_s": 0.917,
+ "env_oct": 1.832,
+ "curve_db": 31.74,
+ "t40_ratio": 2.173,
+ "attack_ratio": 10.08,
+ "active_ratio": 2.061,
+ "ref_median": {
+ "attack_ms": 248.5,
+ "t40_ms": 1587.5,
+ "active_s": 3.189
+ },
+ "phrase_z": 2.77,
+ "phrase_min": 1.81,
+ "margin": 1.36,
+ "wins": 0.5,
+ "best_confusion": "a bass drum",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.467
+ ],
+ [
+ "Explosion",
+ 0.337
+ ],
+ [
+ "Burst, pop",
+ 0.135
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.337,
+ "floor_db": -45.2,
+ "stereo_corr": 0.752,
+ "darken": -2.31,
+ "artifacts": [
+ "late start"
+ ],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -68.0,
+ "first_sample": 0.0,
+ "tail_db": -120.0,
+ "lead_s": 0.138,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity_bench": {
+ "n": 1
+ },
+ "diversity_forge": {
+ "n": 2,
+ "cos_mean": 0.659,
+ "cos_max": 0.659,
+ "tone_min_db": 7.13,
+ "tone_mean_db": 7.13,
+ "near_duplicates": []
+ },
+ "summary": {
+ "made": {},
+ "real_cross": {
+ "n": 2,
+ "tone_db": 6.19,
+ "env_oct": 2.5,
+ "curve_db": 31.98,
+ "phrase_z": 3.04,
+ "margin": 1.09,
+ "as_p": 0.25,
+ "t40_ratio": 0.46,
+ "wins": 0.5,
+ "floor_db": -70.55,
+ "stereo_corr": 0.54,
+ "darken": -1.3,
+ "as_names": 0.5,
+ "artifacts": 0,
+ "var_cos": 0.659
+ },
+ "real_loo": {}
+ }
+ },
+ "logo_sting": {
+ "n_bench": 2,
+ "n_forge": 2,
+ "note": "no real logo stings exist in the library (they are composed); the refs are the struck metal / glass rings a sting ends on: a tail-shape and tone reference only.",
+ "bench_refs": [
+ "E:/earshot/out/refs/logo_sting/bench/vsco-Triangle6-Hit_v1_rr1_1.wav",
+ "E:/earshot/out/refs/logo_sting/bench/fs-170237_1.wav"
+ ],
+ "made": [],
+ "diversity": {
+ "n": 0
+ },
+ "real_cross": [
+ {
+ "file": "E:/earshot/out/refs/logo_sting/forge/kit-sting_tail_01.wav",
+ "length_s": 5.863,
+ "nn_cos": 0.534,
+ "tone_db": 14.94,
+ "tone_ratio": null,
+ "n_outside": 15,
+ "outside": [
+ [
+ "62 Hz",
+ -23.2,
+ "below"
+ ],
+ [
+ "79 Hz",
+ -22.5,
+ "below"
+ ],
+ [
+ "99 Hz",
+ -22.5,
+ "below"
+ ]
+ ],
+ "attack_ms": 5.0,
+ "t40_ms": 2065.0,
+ "active_s": 2.765,
+ "onset_s": 0.005,
+ "env_oct": 1.114,
+ "curve_db": 23.97,
+ "t40_ratio": 2.394,
+ "attack_ratio": 1.429,
+ "active_ratio": 2.967,
+ "ref_median": {
+ "attack_ms": 3.5,
+ "t40_ms": 862.5,
+ "active_s": 0.932
+ },
+ "phrase_z": 2.18,
+ "phrase_min": 0.59,
+ "margin": -0.61,
+ "wins": 0.5,
+ "best_confusion": "a doorbell",
+ "as_top": [
+ [
+ "Bicycle bell",
+ 0.188
+ ],
+ [
+ "Bell",
+ 0.147
+ ],
+ [
+ "Tuning fork",
+ 0.097
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.147,
+ "floor_db": -67.1,
+ "stereo_corr": 1.0,
+ "darken": -1.4,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -125.9,
+ "first_sample": 0.0,
+ "tail_db": -115.1,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/logo_sting/forge/mk-3108_1.wav",
+ "length_s": 1.135,
+ "nn_cos": 0.17,
+ "tone_db": 16.1,
+ "tone_ratio": null,
+ "n_outside": 15,
+ "outside": [
+ [
+ "198 Hz",
+ -26.0,
+ "below"
+ ],
+ [
+ "62 Hz",
+ -23.2,
+ "below"
+ ],
+ [
+ "125 Hz",
+ -23.2,
+ "below"
+ ]
+ ],
+ "attack_ms": 131.0,
+ "t40_ms": 865.0,
+ "active_s": 1.039,
+ "onset_s": 0.036,
+ "env_oct": 1.796,
+ "curve_db": 18.58,
+ "t40_ratio": 1.003,
+ "attack_ratio": 37.429,
+ "active_ratio": 1.115,
+ "ref_median": {
+ "attack_ms": 3.5,
+ "t40_ms": 862.5,
+ "active_s": 0.932
+ },
+ "phrase_z": 0.62,
+ "phrase_min": 0.32,
+ "margin": -1.03,
+ "wins": 0.0,
+ "best_confusion": "a cymbal crash",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.198
+ ],
+ [
+ "Music",
+ 0.053
+ ],
+ [
+ "Ringtone",
+ 0.051
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.003,
+ "floor_db": -37.1,
+ "stereo_corr": 0.187,
+ "darken": -0.3,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -134.8,
+ "first_sample": 0.0,
+ "tail_db": -61.5,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_loo": [
+ {
+ "file": "E:/earshot/out/refs/logo_sting/bench/vsco-Triangle6-Hit_v1_rr1_1.wav",
+ "length_s": 1.065,
+ "nn_cos": 0.507,
+ "tone_db": 23.22,
+ "tone_ratio": null,
+ "n_outside": 27,
+ "outside": [
+ [
+ "794 Hz",
+ -42.0,
+ "below"
+ ],
+ [
+ "1.59 kHz",
+ -41.6,
+ "below"
+ ],
+ [
+ "1 kHz",
+ -37.8,
+ "below"
+ ]
+ ],
+ "attack_ms": 2.0,
+ "t40_ms": 660.0,
+ "active_s": 0.795,
+ "onset_s": 0.0,
+ "env_oct": 0.813,
+ "curve_db": 8.99,
+ "t40_ratio": 0.62,
+ "attack_ratio": 0.4,
+ "active_ratio": 0.744,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 1065.0,
+ "active_s": 1.069
+ },
+ "phrase_z": 1.39,
+ "phrase_min": 0.27,
+ "margin": -0.83,
+ "wins": 0.5,
+ "best_confusion": "a cymbal crash",
+ "as_top": [
+ [
+ "Ding",
+ 0.472
+ ],
+ [
+ "Bicycle bell",
+ 0.113
+ ],
+ [
+ "Clang",
+ 0.068
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.472,
+ "floor_db": -47.9,
+ "stereo_corr": 0.612,
+ "darken": -0.61,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -121.7,
+ "first_sample": 0.0,
+ "tail_db": -66.3,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/logo_sting/bench/fs-170237_1.wav",
+ "length_s": 1.2,
+ "nn_cos": 0.507,
+ "tone_db": 23.22,
+ "tone_ratio": null,
+ "n_outside": 27,
+ "outside": [
+ [
+ "794 Hz",
+ 42.0,
+ "above"
+ ],
+ [
+ "1.59 kHz",
+ 41.6,
+ "above"
+ ],
+ [
+ "1 kHz",
+ 37.8,
+ "above"
+ ]
+ ],
+ "attack_ms": 5.0,
+ "t40_ms": 1065.0,
+ "active_s": 1.069,
+ "onset_s": 0.001,
+ "env_oct": 0.813,
+ "curve_db": 8.99,
+ "t40_ratio": 1.614,
+ "attack_ratio": 2.5,
+ "active_ratio": 1.345,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 660.0,
+ "active_s": 0.795
+ },
+ "phrase_z": 1.33,
+ "phrase_min": -0.26,
+ "margin": -0.33,
+ "wins": 0.5,
+ "best_confusion": "a gong",
+ "as_top": [
+ [
+ "Ding",
+ 0.416
+ ],
+ [
+ "Clang",
+ 0.312
+ ],
+ [
+ "Bell",
+ 0.016
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.416,
+ "floor_db": -40.3,
+ "stereo_corr": 1.0,
+ "darken": -2.26,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -93.8,
+ "first_sample": 0.0,
+ "tail_db": -59.9,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_rev": [
+ {
+ "file": "E:/earshot/out/refs/logo_sting/bench/vsco-Triangle6-Hit_v1_rr1_1.wav",
+ "length_s": 1.065,
+ "nn_cos": 0.5,
+ "tone_db": 15.08,
+ "tone_ratio": null,
+ "n_outside": 20,
+ "outside": [
+ [
+ "1.59 kHz",
+ -26.1,
+ "below"
+ ],
+ [
+ "1.26 kHz",
+ -24.5,
+ "below"
+ ],
+ [
+ "1 kHz",
+ -24.1,
+ "below"
+ ]
+ ],
+ "attack_ms": 2.0,
+ "t40_ms": 660.0,
+ "active_s": 0.795,
+ "onset_s": 0.0,
+ "env_oct": 2.499,
+ "curve_db": 15.5,
+ "t40_ratio": 0.451,
+ "attack_ratio": 0.029,
+ "active_ratio": 0.418,
+ "ref_median": {
+ "attack_ms": 68.0,
+ "t40_ms": 1465.0,
+ "active_s": 1.902
+ },
+ "phrase_z": 1.39,
+ "phrase_min": 0.27,
+ "margin": -0.83,
+ "wins": 0.5,
+ "best_confusion": "a cymbal crash",
+ "as_top": [
+ [
+ "Ding",
+ 0.472
+ ],
+ [
+ "Bicycle bell",
+ 0.113
+ ],
+ [
+ "Clang",
+ 0.068
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.472,
+ "floor_db": -47.9,
+ "stereo_corr": 0.612,
+ "darken": -0.61,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -121.7,
+ "first_sample": 0.0,
+ "tail_db": -66.3,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/logo_sting/bench/fs-170237_1.wav",
+ "length_s": 1.2,
+ "nn_cos": 0.534,
+ "tone_db": 21.33,
+ "tone_ratio": null,
+ "n_outside": 21,
+ "outside": [
+ [
+ "198 Hz",
+ 33.2,
+ "above"
+ ],
+ [
+ "250 Hz",
+ 30.5,
+ "above"
+ ],
+ [
+ "315 Hz",
+ 29.3,
+ "above"
+ ]
+ ],
+ "attack_ms": 5.0,
+ "t40_ms": 1065.0,
+ "active_s": 1.069,
+ "onset_s": 0.001,
+ "env_oct": 1.686,
+ "curve_db": 14.6,
+ "t40_ratio": 0.727,
+ "attack_ratio": 0.074,
+ "active_ratio": 0.562,
+ "ref_median": {
+ "attack_ms": 68.0,
+ "t40_ms": 1465.0,
+ "active_s": 1.902
+ },
+ "phrase_z": 1.33,
+ "phrase_min": -0.26,
+ "margin": -0.33,
+ "wins": 0.5,
+ "best_confusion": "a gong",
+ "as_top": [
+ [
+ "Ding",
+ 0.416
+ ],
+ [
+ "Clang",
+ 0.312
+ ],
+ [
+ "Bell",
+ 0.016
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.416,
+ "floor_db": -40.3,
+ "stereo_corr": 1.0,
+ "darken": -2.26,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -93.8,
+ "first_sample": 0.0,
+ "tail_db": -59.9,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity_bench": {
+ "n": 2,
+ "cos_mean": 0.507,
+ "cos_max": 0.507,
+ "tone_min_db": 23.22,
+ "tone_mean_db": 23.22,
+ "near_duplicates": []
+ },
+ "diversity_forge": {
+ "n": 2,
+ "cos_mean": 0.185,
+ "cos_max": 0.185,
+ "tone_min_db": 11.82,
+ "tone_mean_db": 11.82,
+ "near_duplicates": []
+ },
+ "summary": {
+ "made": {},
+ "real_cross": {
+ "n": 2,
+ "tone_db": 15.52,
+ "env_oct": 1.46,
+ "curve_db": 21.27,
+ "phrase_z": 1.4,
+ "margin": -0.82,
+ "as_p": 0.07,
+ "t40_ratio": 1.7,
+ "wins": 0.25,
+ "floor_db": -52.1,
+ "stereo_corr": 0.59,
+ "darken": -0.85,
+ "as_names": 0.5,
+ "artifacts": 0,
+ "var_cos": 0.185
+ },
+ "real_loo": {
+ "n": 2,
+ "tone_db": 23.22,
+ "env_oct": 0.81,
+ "curve_db": 8.99,
+ "phrase_z": 1.36,
+ "margin": -0.58,
+ "as_p": 0.44,
+ "t40_ratio": 1.12,
+ "wins": 0.5,
+ "floor_db": -44.1,
+ "stereo_corr": 0.81,
+ "darken": -1.43,
+ "as_names": 1.0,
+ "artifacts": 0,
+ "var_cos": 0.507
+ }
+ }
+ },
+ "notify": {
+ "n_bench": 5,
+ "n_forge": 5,
+ "note": "struck bells / glass / kalimba (forge) and celesta / glass / triangle (bench), rings capped at 2 s with a 0.25 s fade; no designed UI dings in the library.",
+ "bench_refs": [
+ "E:/earshot/out/refs/notify/bench/fs-410636_1.wav",
+ "E:/earshot/out/refs/notify/bench/fs-410649_1.wav",
+ "E:/earshot/out/refs/notify/bench/fs-410664_1.wav",
+ "E:/earshot/out/refs/notify/bench/fs-170237_1.wav",
+ "E:/earshot/out/refs/notify/bench/vsco-Triangle6-Hit_v1_rr1_1.wav"
+ ],
+ "made": [],
+ "diversity": {
+ "n": 0
+ },
+ "real_cross": [
+ {
+ "file": "E:/earshot/out/refs/notify/forge/kit-ui_notify_01.wav",
+ "length_s": 1.6,
+ "typ_z": -0.24,
+ "nn_cos": 0.543,
+ "tone_db": 21.3,
+ "tone_ratio": 1.28,
+ "n_outside": 13,
+ "outside": [
+ [
+ "4 kHz",
+ 40.3,
+ "above"
+ ],
+ [
+ "500 Hz",
+ -31.3,
+ "below"
+ ],
+ [
+ "397 Hz",
+ -25.2,
+ "below"
+ ]
+ ],
+ "attack_ms": 58.0,
+ "t40_ms": 475.0,
+ "active_s": 0.733,
+ "onset_s": 0.002,
+ "env_oct": 2.124,
+ "curve_db": 30.33,
+ "t40_ratio": 0.266,
+ "attack_ratio": 8.286,
+ "active_ratio": 0.376,
+ "ref_median": {
+ "attack_ms": 7.0,
+ "t40_ms": 1785.0,
+ "active_s": 1.949
+ },
+ "phrase_z": 2.24,
+ "phrase_min": 0.88,
+ "margin": -0.05,
+ "wins": 0.5,
+ "best_confusion": "a doorbell",
+ "as_top": [
+ [
+ "Ding",
+ 0.26
+ ],
+ [
+ "Bicycle bell",
+ 0.042
+ ],
+ [
+ "Ping",
+ 0.037
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.26,
+ "floor_db": -82.3,
+ "stereo_corr": 1.0,
+ "darken": -0.18,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -142.1,
+ "first_sample": 0.0,
+ "tail_db": -105.5,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/notify/forge/fs-681787_1.wav",
+ "length_s": 1.745,
+ "typ_z": -0.61,
+ "nn_cos": 0.478,
+ "tone_db": 17.36,
+ "tone_ratio": 1.04,
+ "n_outside": 7,
+ "outside": [
+ [
+ "397 Hz",
+ 34.4,
+ "above"
+ ],
+ [
+ "50 Hz",
+ 26.6,
+ "above"
+ ],
+ [
+ "79 Hz",
+ 23.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 7.0,
+ "t40_ms": 1510.0,
+ "active_s": 1.529,
+ "onset_s": 0.006,
+ "env_oct": 0.197,
+ "curve_db": 12.46,
+ "t40_ratio": 0.846,
+ "attack_ratio": 1.0,
+ "active_ratio": 0.785,
+ "ref_median": {
+ "attack_ms": 7.0,
+ "t40_ms": 1785.0,
+ "active_s": 1.949
+ },
+ "phrase_z": 0.44,
+ "phrase_min": -0.63,
+ "margin": -0.43,
+ "wins": 0.0,
+ "best_confusion": "a doorbell",
+ "as_top": [
+ [
+ "Ding",
+ 0.525
+ ],
+ [
+ "Clang",
+ 0.405
+ ],
+ [
+ "Sound effect",
+ 0.045
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.525,
+ "floor_db": -41.2,
+ "stereo_corr": 0.986,
+ "darken": -1.57,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -83.4,
+ "first_sample": 0.0,
+ "tail_db": -60.0,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/notify/forge/fs-681789_1.wav",
+ "length_s": 1.165,
+ "typ_z": 0.34,
+ "nn_cos": 0.63,
+ "tone_db": 13.32,
+ "tone_ratio": 0.8,
+ "n_outside": 6,
+ "outside": [
+ [
+ "500 Hz",
+ 28.5,
+ "above"
+ ],
+ [
+ "50 Hz",
+ 27.0,
+ "above"
+ ],
+ [
+ "62 Hz",
+ 21.7,
+ "above"
+ ]
+ ],
+ "attack_ms": 11.0,
+ "t40_ms": 1005.0,
+ "active_s": 1.031,
+ "onset_s": 0.009,
+ "env_oct": 0.8,
+ "curve_db": 24.07,
+ "t40_ratio": 0.563,
+ "attack_ratio": 1.571,
+ "active_ratio": 0.529,
+ "ref_median": {
+ "attack_ms": 7.0,
+ "t40_ms": 1785.0,
+ "active_s": 1.949
+ },
+ "phrase_z": 1.25,
+ "phrase_min": 0.41,
+ "margin": -0.24,
+ "wins": 0.0,
+ "best_confusion": "a doorbell",
+ "as_top": [
+ [
+ "Clang",
+ 0.234
+ ],
+ [
+ "Ding",
+ 0.216
+ ],
+ [
+ "Sound effect",
+ 0.075
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.216,
+ "floor_db": -40.7,
+ "stereo_corr": 0.995,
+ "darken": -0.31,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -88.3,
+ "first_sample": 0.0,
+ "tail_db": -59.9,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/notify/forge/fs-681792_1.wav",
+ "length_s": 1.36,
+ "typ_z": -0.66,
+ "nn_cos": 0.481,
+ "tone_db": 24.72,
+ "tone_ratio": 1.48,
+ "n_outside": 10,
+ "outside": [
+ [
+ "50 Hz",
+ 35.7,
+ "above"
+ ],
+ [
+ "39 Hz",
+ 31.3,
+ "above"
+ ],
+ [
+ "62 Hz",
+ 30.9,
+ "above"
+ ]
+ ],
+ "attack_ms": 8.0,
+ "t40_ms": 460.0,
+ "active_s": 0.576,
+ "onset_s": 0.004,
+ "env_oct": 1.302,
+ "curve_db": 29.02,
+ "t40_ratio": 0.258,
+ "attack_ratio": 1.143,
+ "active_ratio": 0.296,
+ "ref_median": {
+ "attack_ms": 7.0,
+ "t40_ms": 1785.0,
+ "active_s": 1.949
+ },
+ "phrase_z": 0.34,
+ "phrase_min": -0.8,
+ "margin": -0.9,
+ "wins": 0.0,
+ "best_confusion": "a church bell",
+ "as_top": [
+ [
+ "Clang",
+ 0.639
+ ],
+ [
+ "Ding",
+ 0.302
+ ],
+ [
+ "Cowbell",
+ 0.043
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.302,
+ "floor_db": -61.1,
+ "stereo_corr": 0.893,
+ "darken": -1.42,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -81.9,
+ "first_sample": 0.0,
+ "tail_db": -62.6,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/notify/forge/fs-549899_1.wav",
+ "length_s": 0.83,
+ "typ_z": 0.26,
+ "nn_cos": 0.848,
+ "tone_db": 23.8,
+ "tone_ratio": 1.42,
+ "n_outside": 13,
+ "outside": [
+ [
+ "4 kHz",
+ 39.4,
+ "above"
+ ],
+ [
+ "500 Hz",
+ -28.6,
+ "below"
+ ],
+ [
+ "397 Hz",
+ -21.9,
+ "below"
+ ]
+ ],
+ "attack_ms": 2.0,
+ "t40_ms": 610.0,
+ "active_s": 0.79,
+ "onset_s": 0.0,
+ "env_oct": 1.553,
+ "curve_db": 28.92,
+ "t40_ratio": 0.342,
+ "attack_ratio": 0.286,
+ "active_ratio": 0.405,
+ "ref_median": {
+ "attack_ms": 7.0,
+ "t40_ms": 1785.0,
+ "active_s": 1.949
+ },
+ "phrase_z": 1.93,
+ "phrase_min": 0.96,
+ "margin": 0.13,
+ "wins": 0.5,
+ "best_confusion": "a church bell",
+ "as_top": [
+ [
+ "Ding",
+ 0.437
+ ],
+ [
+ "Bicycle bell",
+ 0.074
+ ],
+ [
+ "Bell",
+ 0.033
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.437,
+ "floor_db": -38.2,
+ "stereo_corr": 1.0,
+ "darken": -1.06,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -128.2,
+ "first_sample": 0.0,
+ "tail_db": -63.8,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_loo": [
+ {
+ "file": "E:/earshot/out/refs/notify/bench/fs-410636_1.wav",
+ "length_s": 2.0,
+ "typ_z": 4.08,
+ "nn_cos": 0.905,
+ "tone_db": 14.51,
+ "tone_ratio": 0.7,
+ "n_outside": 2,
+ "outside": [
+ [
+ "630 Hz",
+ 33.2,
+ "above"
+ ],
+ [
+ "5.04 kHz",
+ -30.1,
+ "below"
+ ]
+ ],
+ "attack_ms": 10.0,
+ "t40_ms": 1950.0,
+ "active_s": 1.958,
+ "onset_s": 0.007,
+ "env_oct": 0.522,
+ "curve_db": 15.49,
+ "t40_ratio": 1.368,
+ "attack_ratio": 1.667,
+ "active_ratio": 1.298,
+ "ref_median": {
+ "attack_ms": 6.0,
+ "t40_ms": 1425.0,
+ "active_s": 1.509
+ },
+ "phrase_z": 2.45,
+ "phrase_min": 1.89,
+ "margin": -0.97,
+ "wins": 0.0,
+ "best_confusion": "a doorbell",
+ "as_top": [
+ [
+ "Doorbell",
+ 0.254
+ ],
+ [
+ "Ding",
+ 0.148
+ ],
+ [
+ "Ding-dong",
+ 0.042
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.148,
+ "floor_db": -23.3,
+ "stereo_corr": -0.687,
+ "darken": -0.53,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -96.5,
+ "first_sample": 0.0,
+ "tail_db": -56.2,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/notify/bench/fs-410649_1.wav",
+ "length_s": 2.0,
+ "typ_z": 5.0,
+ "nn_cos": 0.905,
+ "tone_db": 13.76,
+ "tone_ratio": 0.66,
+ "n_outside": 2,
+ "outside": [
+ [
+ "794 Hz",
+ 36.5,
+ "above"
+ ],
+ [
+ "31 Hz",
+ 9.4,
+ "above"
+ ]
+ ],
+ "attack_ms": 22.0,
+ "t40_ms": 1785.0,
+ "active_s": 1.949,
+ "onset_s": 0.021,
+ "env_oct": 0.83,
+ "curve_db": 17.41,
+ "t40_ratio": 1.186,
+ "attack_ratio": 3.667,
+ "active_ratio": 1.291,
+ "ref_median": {
+ "attack_ms": 6.0,
+ "t40_ms": 1505.0,
+ "active_s": 1.51
+ },
+ "phrase_z": 2.04,
+ "phrase_min": 1.47,
+ "margin": -0.91,
+ "wins": 0.0,
+ "best_confusion": "a doorbell",
+ "as_top": [
+ [
+ "Doorbell",
+ 0.385
+ ],
+ [
+ "Ding",
+ 0.23
+ ],
+ [
+ "Ding-dong",
+ 0.082
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.23,
+ "floor_db": -22.1,
+ "stereo_corr": -0.989,
+ "darken": -0.27,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -92.9,
+ "first_sample": 0.0,
+ "tail_db": -54.2,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/notify/bench/fs-410664_1.wav",
+ "length_s": 2.0,
+ "typ_z": 3.84,
+ "nn_cos": 0.894,
+ "tone_db": 16.7,
+ "tone_ratio": 0.85,
+ "n_outside": 13,
+ "outside": [
+ [
+ "1 kHz",
+ 38.3,
+ "above"
+ ],
+ [
+ "4 kHz",
+ -26.4,
+ "below"
+ ],
+ [
+ "2.52 kHz",
+ -18.8,
+ "below"
+ ]
+ ],
+ "attack_ms": 7.0,
+ "t40_ms": 1945.0,
+ "active_s": 1.951,
+ "onset_s": 0.009,
+ "env_oct": 0.306,
+ "curve_db": 15.6,
+ "t40_ratio": 1.365,
+ "attack_ratio": 0.933,
+ "active_ratio": 1.293,
+ "ref_median": {
+ "attack_ms": 7.5,
+ "t40_ms": 1425.0,
+ "active_s": 1.509
+ },
+ "phrase_z": 1.74,
+ "phrase_min": 1.14,
+ "margin": -0.94,
+ "wins": 0.0,
+ "best_confusion": "a doorbell",
+ "as_top": [
+ [
+ "Ding",
+ 0.366
+ ],
+ [
+ "Doorbell",
+ 0.083
+ ],
+ [
+ "Clang",
+ 0.065
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.366,
+ "floor_db": -24.4,
+ "stereo_corr": 0.721,
+ "darken": -0.07,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -90.1,
+ "first_sample": 0.0,
+ "tail_db": -57.1,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/notify/bench/fs-170237_1.wav",
+ "length_s": 1.2,
+ "typ_z": -0.96,
+ "nn_cos": 0.507,
+ "tone_db": 28.68,
+ "tone_ratio": 1.84,
+ "n_outside": 20,
+ "outside": [
+ [
+ "6.35 kHz",
+ 46.8,
+ "above"
+ ],
+ [
+ "16 kHz",
+ 45.4,
+ "above"
+ ],
+ [
+ "12.7 kHz",
+ 42.4,
+ "above"
+ ]
+ ],
+ "attack_ms": 5.0,
+ "t40_ms": 1065.0,
+ "active_s": 1.069,
+ "onset_s": 0.001,
+ "env_oct": 0.814,
+ "curve_db": 26.44,
+ "t40_ratio": 0.571,
+ "attack_ratio": 0.588,
+ "active_ratio": 0.548,
+ "ref_median": {
+ "attack_ms": 8.5,
+ "t40_ms": 1865.0,
+ "active_s": 1.95
+ },
+ "phrase_z": 1.33,
+ "phrase_min": 0.3,
+ "margin": -0.24,
+ "wins": 0.0,
+ "best_confusion": "a church bell",
+ "as_top": [
+ [
+ "Ding",
+ 0.416
+ ],
+ [
+ "Clang",
+ 0.312
+ ],
+ [
+ "Bell",
+ 0.016
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.416,
+ "floor_db": -40.3,
+ "stereo_corr": 1.0,
+ "darken": -2.26,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -93.8,
+ "first_sample": 0.0,
+ "tail_db": -59.9,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/notify/bench/vsco-Triangle6-Hit_v1_rr1_1.wav",
+ "length_s": 1.065,
+ "typ_z": -1.59,
+ "nn_cos": 0.507,
+ "tone_db": 24.79,
+ "tone_ratio": 1.59,
+ "n_outside": 9,
+ "outside": [
+ [
+ "8 kHz",
+ 52.1,
+ "above"
+ ],
+ [
+ "5.04 kHz",
+ 48.9,
+ "above"
+ ],
+ [
+ "630 Hz",
+ -31.0,
+ "below"
+ ]
+ ],
+ "attack_ms": 2.0,
+ "t40_ms": 660.0,
+ "active_s": 0.795,
+ "onset_s": 0.0,
+ "env_oct": 1.627,
+ "curve_db": 28.67,
+ "t40_ratio": 0.354,
+ "attack_ratio": 0.235,
+ "active_ratio": 0.408,
+ "ref_median": {
+ "attack_ms": 8.5,
+ "t40_ms": 1865.0,
+ "active_s": 1.95
+ },
+ "phrase_z": 1.76,
+ "phrase_min": 1.43,
+ "margin": 0.2,
+ "wins": 0.0,
+ "best_confusion": "a doorbell",
+ "as_top": [
+ [
+ "Ding",
+ 0.472
+ ],
+ [
+ "Bicycle bell",
+ 0.113
+ ],
+ [
+ "Clang",
+ 0.068
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.472,
+ "floor_db": -47.9,
+ "stereo_corr": 0.612,
+ "darken": -0.61,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -121.7,
+ "first_sample": 0.0,
+ "tail_db": -66.3,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_rev": [
+ {
+ "file": "E:/earshot/out/refs/notify/bench/fs-410636_1.wav",
+ "length_s": 2.0,
+ "typ_z": -1.79,
+ "nn_cos": 0.593,
+ "tone_db": 19.93,
+ "tone_ratio": 0.89,
+ "n_outside": 6,
+ "outside": [
+ [
+ "8 kHz",
+ -38.4,
+ "below"
+ ],
+ [
+ "630 Hz",
+ 37.0,
+ "above"
+ ],
+ [
+ "4 kHz",
+ -34.5,
+ "below"
+ ]
+ ],
+ "attack_ms": 10.0,
+ "t40_ms": 1950.0,
+ "active_s": 1.958,
+ "onset_s": 0.007,
+ "env_oct": 1.103,
+ "curve_db": 32.82,
+ "t40_ratio": 3.197,
+ "attack_ratio": 1.25,
+ "active_ratio": 2.478,
+ "ref_median": {
+ "attack_ms": 8.0,
+ "t40_ms": 610.0,
+ "active_s": 0.79
+ },
+ "phrase_z": 2.45,
+ "phrase_min": 1.89,
+ "margin": -0.97,
+ "wins": 0.0,
+ "best_confusion": "a doorbell",
+ "as_top": [
+ [
+ "Doorbell",
+ 0.254
+ ],
+ [
+ "Ding",
+ 0.148
+ ],
+ [
+ "Ding-dong",
+ 0.042
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.148,
+ "floor_db": -23.3,
+ "stereo_corr": -0.687,
+ "darken": -0.53,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -96.5,
+ "first_sample": 0.0,
+ "tail_db": -56.2,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/notify/bench/fs-410649_1.wav",
+ "length_s": 2.0,
+ "typ_z": -1.55,
+ "nn_cos": 0.63,
+ "tone_db": 17.36,
+ "tone_ratio": 0.77,
+ "n_outside": 7,
+ "outside": [
+ [
+ "8 kHz",
+ -42.5,
+ "below"
+ ],
+ [
+ "794 Hz",
+ 36.2,
+ "above"
+ ],
+ [
+ "6.35 kHz",
+ -27.6,
+ "below"
+ ]
+ ],
+ "attack_ms": 22.0,
+ "t40_ms": 1785.0,
+ "active_s": 1.949,
+ "onset_s": 0.021,
+ "env_oct": 1.437,
+ "curve_db": 33.82,
+ "t40_ratio": 2.926,
+ "attack_ratio": 2.75,
+ "active_ratio": 2.467,
+ "ref_median": {
+ "attack_ms": 8.0,
+ "t40_ms": 610.0,
+ "active_s": 0.79
+ },
+ "phrase_z": 2.04,
+ "phrase_min": 1.47,
+ "margin": -0.91,
+ "wins": 0.0,
+ "best_confusion": "a doorbell",
+ "as_top": [
+ [
+ "Doorbell",
+ 0.385
+ ],
+ [
+ "Ding",
+ 0.23
+ ],
+ [
+ "Ding-dong",
+ 0.082
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.23,
+ "floor_db": -22.1,
+ "stereo_corr": -0.989,
+ "darken": -0.27,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -92.9,
+ "first_sample": 0.0,
+ "tail_db": -54.2,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/notify/bench/fs-410664_1.wav",
+ "length_s": 2.0,
+ "typ_z": -1.66,
+ "nn_cos": 0.616,
+ "tone_db": 23.76,
+ "tone_ratio": 1.06,
+ "n_outside": 7,
+ "outside": [
+ [
+ "4 kHz",
+ -45.2,
+ "below"
+ ],
+ [
+ "8 kHz",
+ -42.5,
+ "below"
+ ],
+ [
+ "1 kHz",
+ 37.7,
+ "above"
+ ]
+ ],
+ "attack_ms": 7.0,
+ "t40_ms": 1945.0,
+ "active_s": 1.951,
+ "onset_s": 0.009,
+ "env_oct": 1.057,
+ "curve_db": 33.0,
+ "t40_ratio": 3.189,
+ "attack_ratio": 0.875,
+ "active_ratio": 2.47,
+ "ref_median": {
+ "attack_ms": 8.0,
+ "t40_ms": 610.0,
+ "active_s": 0.79
+ },
+ "phrase_z": 1.74,
+ "phrase_min": 1.14,
+ "margin": -0.94,
+ "wins": 0.0,
+ "best_confusion": "a doorbell",
+ "as_top": [
+ [
+ "Ding",
+ 0.366
+ ],
+ [
+ "Doorbell",
+ 0.083
+ ],
+ [
+ "Clang",
+ 0.065
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.366,
+ "floor_db": -24.4,
+ "stereo_corr": 0.721,
+ "darken": -0.07,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -90.1,
+ "first_sample": 0.0,
+ "tail_db": -57.1,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/notify/bench/fs-170237_1.wav",
+ "length_s": 1.2,
+ "typ_z": -0.42,
+ "nn_cos": 0.848,
+ "tone_db": 16.61,
+ "tone_ratio": 0.74,
+ "n_outside": 12,
+ "outside": [
+ [
+ "16 kHz",
+ 32.6,
+ "above"
+ ],
+ [
+ "10.1 kHz",
+ 26.3,
+ "above"
+ ],
+ [
+ "794 Hz",
+ 24.4,
+ "above"
+ ]
+ ],
+ "attack_ms": 5.0,
+ "t40_ms": 1065.0,
+ "active_s": 1.069,
+ "onset_s": 0.001,
+ "env_oct": 0.639,
+ "curve_db": 9.46,
+ "t40_ratio": 1.746,
+ "attack_ratio": 0.625,
+ "active_ratio": 1.353,
+ "ref_median": {
+ "attack_ms": 8.0,
+ "t40_ms": 610.0,
+ "active_s": 0.79
+ },
+ "phrase_z": 1.33,
+ "phrase_min": 0.3,
+ "margin": -0.24,
+ "wins": 0.0,
+ "best_confusion": "a church bell",
+ "as_top": [
+ [
+ "Ding",
+ 0.416
+ ],
+ [
+ "Clang",
+ 0.312
+ ],
+ [
+ "Bell",
+ 0.016
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.416,
+ "floor_db": -40.3,
+ "stereo_corr": 1.0,
+ "darken": -2.26,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -93.8,
+ "first_sample": 0.0,
+ "tail_db": -59.9,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/notify/bench/vsco-Triangle6-Hit_v1_rr1_1.wav",
+ "length_s": 1.065,
+ "typ_z": -1.54,
+ "nn_cos": 0.576,
+ "tone_db": 18.63,
+ "tone_ratio": 0.83,
+ "n_outside": 6,
+ "outside": [
+ [
+ "5.04 kHz",
+ 31.0,
+ "above"
+ ],
+ [
+ "10.1 kHz",
+ 29.0,
+ "above"
+ ],
+ [
+ "1.59 kHz",
+ -22.6,
+ "below"
+ ]
+ ],
+ "attack_ms": 2.0,
+ "t40_ms": 660.0,
+ "active_s": 0.795,
+ "onset_s": 0.0,
+ "env_oct": 0.708,
+ "curve_db": 5.33,
+ "t40_ratio": 1.082,
+ "attack_ratio": 0.25,
+ "active_ratio": 1.006,
+ "ref_median": {
+ "attack_ms": 8.0,
+ "t40_ms": 610.0,
+ "active_s": 0.79
+ },
+ "phrase_z": 1.76,
+ "phrase_min": 1.43,
+ "margin": 0.2,
+ "wins": 0.0,
+ "best_confusion": "a doorbell",
+ "as_top": [
+ [
+ "Ding",
+ 0.472
+ ],
+ [
+ "Bicycle bell",
+ 0.113
+ ],
+ [
+ "Clang",
+ 0.068
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.472,
+ "floor_db": -47.9,
+ "stereo_corr": 0.612,
+ "darken": -0.61,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -121.7,
+ "first_sample": 0.0,
+ "tail_db": -66.3,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity_bench": {
+ "n": 5,
+ "cos_mean": 0.476,
+ "cos_max": 0.905,
+ "tone_min_db": 11.33,
+ "tone_mean_db": 22.84,
+ "near_duplicates": []
+ },
+ "diversity_forge": {
+ "n": 5,
+ "cos_mean": 0.593,
+ "cos_max": 0.869,
+ "tone_min_db": 10.92,
+ "tone_mean_db": 23.48,
+ "near_duplicates": []
+ },
+ "summary": {
+ "made": {},
+ "real_cross": {
+ "n": 5,
+ "typ_z": -0.24,
+ "tone_db": 21.3,
+ "env_oct": 1.3,
+ "curve_db": 28.92,
+ "phrase_z": 1.25,
+ "margin": -0.24,
+ "as_p": 0.3,
+ "tone_ratio": 1.28,
+ "t40_ratio": 0.34,
+ "wins": 0.0,
+ "floor_db": -41.2,
+ "stereo_corr": 0.99,
+ "darken": -1.06,
+ "as_names": 1.0,
+ "artifacts": 0,
+ "var_cos": 0.593
+ },
+ "real_loo": {
+ "n": 5,
+ "typ_z": 3.84,
+ "tone_db": 16.7,
+ "env_oct": 0.81,
+ "curve_db": 17.41,
+ "phrase_z": 1.76,
+ "margin": -0.91,
+ "as_p": 0.37,
+ "tone_ratio": 0.85,
+ "t40_ratio": 1.19,
+ "wins": 0.0,
+ "floor_db": -24.4,
+ "stereo_corr": 0.61,
+ "darken": -0.53,
+ "as_names": 1.0,
+ "artifacts": 0,
+ "var_cos": 0.476
+ }
+ }
+ },
+ "pop": {
+ "n_bench": 6,
+ "n_forge": 7,
+ "note": "",
+ "bench_refs": [
+ "E:/earshot/out/refs/pop/bench/fs-202230_1.wav",
+ "E:/earshot/out/refs/pop/bench/fs-33369_1.wav",
+ "E:/earshot/out/refs/pop/bench/fs-545201_1.wav",
+ "E:/earshot/out/refs/pop/bench/fs-573152_1.wav",
+ "E:/earshot/out/refs/pop/bench/mk-3192_1.wav",
+ "E:/earshot/out/refs/pop/bench/mk-3070_1.wav"
+ ],
+ "made": [],
+ "diversity": {
+ "n": 0
+ },
+ "real_cross": [
+ {
+ "file": "E:/earshot/out/refs/pop/forge/kit-ui_pop_01.wav",
+ "length_s": 0.19,
+ "typ_z": 2.12,
+ "nn_cos": 0.933,
+ "tone_db": 11.16,
+ "tone_ratio": 0.82,
+ "n_outside": 3,
+ "outside": [
+ [
+ "1.59 kHz",
+ -23.5,
+ "below"
+ ],
+ [
+ "2 kHz",
+ -18.3,
+ "below"
+ ],
+ [
+ "99 Hz",
+ 12.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 6.0,
+ "t40_ms": 35.0,
+ "active_s": 0.101,
+ "onset_s": 0.009,
+ "env_oct": 1.199,
+ "curve_db": 10.63,
+ "t40_ratio": 0.318,
+ "attack_ratio": 3.0,
+ "active_ratio": 0.78,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 110.0,
+ "active_s": 0.13
+ },
+ "phrase_z": 2.81,
+ "phrase_min": 2.55,
+ "margin": -0.91,
+ "wins": 0.0,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Throat clearing",
+ 0.26
+ ],
+ [
+ "Speech",
+ 0.116
+ ],
+ [
+ "Sound effect",
+ 0.065
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.015,
+ "floor_db": -56.0,
+ "stereo_corr": 1.0,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -84.9,
+ "first_sample": 0.0,
+ "tail_db": -58.6,
+ "lead_s": 0.005,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/forge/kit-ui_pop_02.wav",
+ "length_s": 0.3,
+ "typ_z": 0.54,
+ "nn_cos": 0.722,
+ "tone_db": 11.15,
+ "tone_ratio": 0.82,
+ "n_outside": 3,
+ "outside": [
+ [
+ "2.52 kHz",
+ 28.7,
+ "above"
+ ],
+ [
+ "3.17 kHz",
+ 27.6,
+ "above"
+ ],
+ [
+ "2 kHz",
+ 26.8,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 145.0,
+ "active_s": 0.187,
+ "onset_s": 0.008,
+ "env_oct": 0.643,
+ "curve_db": 9.92,
+ "t40_ratio": 1.318,
+ "attack_ratio": 0.5,
+ "active_ratio": 1.444,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 110.0,
+ "active_s": 0.13
+ },
+ "phrase_z": 2.47,
+ "phrase_min": 2.37,
+ "margin": 0.07,
+ "wins": 0.5,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Whip",
+ 0.599
+ ],
+ [
+ "Slap, smack",
+ 0.101
+ ],
+ [
+ "Finger snapping",
+ 0.021
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.004,
+ "floor_db": -46.9,
+ "stereo_corr": 1.0,
+ "darken": 0.04,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -81.7,
+ "first_sample": 0.0,
+ "tail_db": -56.9,
+ "lead_s": 0.005,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/forge/kit-ui_pop_03.wav",
+ "length_s": 0.414,
+ "typ_z": 0.32,
+ "nn_cos": 0.706,
+ "tone_db": 9.4,
+ "tone_ratio": 0.69,
+ "n_outside": 3,
+ "outside": [
+ [
+ "16 kHz",
+ -14.8,
+ "below"
+ ],
+ [
+ "12.7 kHz",
+ -14.4,
+ "below"
+ ],
+ [
+ "500 Hz",
+ 11.8,
+ "above"
+ ]
+ ],
+ "attack_ms": 52.0,
+ "t40_ms": 100.0,
+ "active_s": 0.168,
+ "onset_s": 0.057,
+ "env_oct": 1.738,
+ "curve_db": 26.9,
+ "t40_ratio": 0.909,
+ "attack_ratio": 26.0,
+ "active_ratio": 1.297,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 110.0,
+ "active_s": 0.13
+ },
+ "phrase_z": 2.64,
+ "phrase_min": 2.3,
+ "margin": 0.61,
+ "wins": 0.5,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Plop",
+ 0.487
+ ],
+ [
+ "Sound effect",
+ 0.129
+ ],
+ [
+ "Drip",
+ 0.073
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.487,
+ "floor_db": -62.3,
+ "stereo_corr": 1.0,
+ "darken": 0.38,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -66.0,
+ "first_sample": 0.0,
+ "tail_db": -82.3,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/forge/kit-ui_pop_04.wav",
+ "length_s": 0.236,
+ "typ_z": -1.91,
+ "nn_cos": 0.428,
+ "tone_db": 21.67,
+ "tone_ratio": 1.59,
+ "n_outside": 17,
+ "outside": [
+ [
+ "5.04 kHz",
+ 33.3,
+ "above"
+ ],
+ [
+ "6.35 kHz",
+ 31.3,
+ "above"
+ ],
+ [
+ "198 Hz",
+ 30.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 20.0,
+ "t40_ms": 100.0,
+ "active_s": 0.119,
+ "onset_s": 0.011,
+ "env_oct": 1.194,
+ "curve_db": 9.76,
+ "t40_ratio": 0.909,
+ "attack_ratio": 10.0,
+ "active_ratio": 0.919,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 110.0,
+ "active_s": 0.13
+ },
+ "phrase_z": 0.68,
+ "phrase_min": -0.01,
+ "margin": -1.62,
+ "wins": 0.0,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Chink, clink",
+ 0.059
+ ],
+ [
+ "Clang",
+ 0.048
+ ],
+ [
+ "Dishes, pots, and pans",
+ 0.046
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.004,
+ "floor_db": -92.2,
+ "stereo_corr": 1.0,
+ "darken": -0.35,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -85.5,
+ "first_sample": 0.0,
+ "tail_db": -114.4,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/forge/mk-3000_1.wav",
+ "length_s": 0.5,
+ "typ_z": -0.46,
+ "nn_cos": 0.604,
+ "tone_db": 16.22,
+ "tone_ratio": 1.19,
+ "n_outside": 10,
+ "outside": [
+ [
+ "198 Hz",
+ 31.6,
+ "above"
+ ],
+ [
+ "1 kHz",
+ -28.8,
+ "below"
+ ],
+ [
+ "315 Hz",
+ 24.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 46.0,
+ "t40_ms": 315.0,
+ "active_s": 0.377,
+ "onset_s": 0.008,
+ "env_oct": 2.528,
+ "curve_db": 27.14,
+ "t40_ratio": 2.864,
+ "attack_ratio": 23.0,
+ "active_ratio": 2.911,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 110.0,
+ "active_s": 0.13
+ },
+ "phrase_z": 2.46,
+ "phrase_min": 1.59,
+ "margin": 1.08,
+ "wins": 0.5,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Plop",
+ 0.713
+ ],
+ [
+ "Sound effect",
+ 0.198
+ ],
+ [
+ "Drip",
+ 0.079
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.713,
+ "floor_db": -53.2,
+ "stereo_corr": 0.954,
+ "darken": 1.43,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -84.8,
+ "first_sample": 0.0,
+ "tail_db": -78.6,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/forge/fs-543200_1.wav",
+ "length_s": 0.048,
+ "typ_z": -0.09,
+ "nn_cos": 0.639,
+ "tone_db": 16.06,
+ "tone_ratio": 1.18,
+ "n_outside": 5,
+ "outside": [
+ [
+ "6.35 kHz",
+ 37.5,
+ "above"
+ ],
+ [
+ "5.04 kHz",
+ 34.3,
+ "above"
+ ],
+ [
+ "2.52 kHz",
+ 23.9,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 35.0,
+ "active_s": 0.035,
+ "onset_s": 0.005,
+ "env_oct": 1.513,
+ "curve_db": 20.54,
+ "t40_ratio": 0.318,
+ "attack_ratio": 0.5,
+ "active_ratio": 0.27,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 110.0,
+ "active_s": 0.13
+ },
+ "phrase_z": 1.72,
+ "phrase_min": 1.61,
+ "margin": -0.5,
+ "wins": 0.0,
+ "best_confusion": "a hand clap",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.071
+ ],
+ [
+ "Music",
+ 0.042
+ ],
+ [
+ "Speech",
+ 0.034
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.004,
+ "floor_db": -44.1,
+ "stereo_corr": 1.0,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -64.5,
+ "first_sample": 0.0,
+ "tail_db": -43.2,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/forge/fs-543200_2.wav",
+ "length_s": 0.049,
+ "typ_z": -0.39,
+ "nn_cos": 0.609,
+ "tone_db": 16.06,
+ "tone_ratio": 1.18,
+ "n_outside": 5,
+ "outside": [
+ [
+ "6.35 kHz",
+ 41.4,
+ "above"
+ ],
+ [
+ "5.04 kHz",
+ 32.4,
+ "above"
+ ],
+ [
+ "8 kHz",
+ 32.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 30.0,
+ "active_s": 0.03,
+ "onset_s": 0.005,
+ "env_oct": 1.661,
+ "curve_db": 20.71,
+ "t40_ratio": 0.273,
+ "attack_ratio": 0.5,
+ "active_ratio": 0.232,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 110.0,
+ "active_s": 0.13
+ },
+ "phrase_z": 1.55,
+ "phrase_min": 1.27,
+ "margin": -0.73,
+ "wins": 0.0,
+ "best_confusion": "a hand clap",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.07
+ ],
+ [
+ "Music",
+ 0.043
+ ],
+ [
+ "Speech",
+ 0.034
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.004,
+ "floor_db": -42.4,
+ "stereo_corr": 1.0,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -62.7,
+ "first_sample": 0.0,
+ "tail_db": -46.2,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_loo": [
+ {
+ "file": "E:/earshot/out/refs/pop/bench/fs-202230_1.wav",
+ "length_s": 0.161,
+ "typ_z": -0.66,
+ "nn_cos": 0.642,
+ "tone_db": 14.83,
+ "tone_ratio": 1.1,
+ "n_outside": 7,
+ "outside": [
+ [
+ "157 Hz",
+ -26.6,
+ "below"
+ ],
+ [
+ "62 Hz",
+ -26.2,
+ "below"
+ ],
+ [
+ "79 Hz",
+ -23.7,
+ "below"
+ ]
+ ],
+ "attack_ms": 2.0,
+ "t40_ms": 140.0,
+ "active_s": 0.139,
+ "onset_s": 0.006,
+ "env_oct": 0.213,
+ "curve_db": 13.12,
+ "t40_ratio": 1.4,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.112,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 100.0,
+ "active_s": 0.125
+ },
+ "phrase_z": 1.16,
+ "phrase_min": 1.03,
+ "margin": -1.14,
+ "wins": 0.0,
+ "best_confusion": "a hand clap",
+ "as_top": [
+ [
+ "Plop",
+ 0.159
+ ],
+ [
+ "Boing",
+ 0.131
+ ],
+ [
+ "Frog",
+ 0.102
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.159,
+ "floor_db": -41.9,
+ "stereo_corr": 1.0,
+ "darken": -0.25,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -116.1,
+ "first_sample": 0.0,
+ "tail_db": -48.0,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/bench/fs-33369_1.wav",
+ "length_s": 0.159,
+ "typ_z": 3.39,
+ "nn_cos": 0.832,
+ "tone_db": 13.46,
+ "tone_ratio": 0.97,
+ "n_outside": 5,
+ "outside": [
+ [
+ "1 kHz",
+ -23.5,
+ "below"
+ ],
+ [
+ "39 Hz",
+ 15.5,
+ "above"
+ ],
+ [
+ "157 Hz",
+ 11.4,
+ "above"
+ ]
+ ],
+ "attack_ms": 5.0,
+ "t40_ms": 100.0,
+ "active_s": 0.125,
+ "onset_s": 0.005,
+ "env_oct": 0.562,
+ "curve_db": 13.44,
+ "t40_ratio": 0.833,
+ "attack_ratio": 2.5,
+ "active_ratio": 0.933,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 120.0,
+ "active_s": 0.134
+ },
+ "phrase_z": 2.89,
+ "phrase_min": 2.72,
+ "margin": -0.46,
+ "wins": 0.0,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.141
+ ],
+ [
+ "Speech",
+ 0.094
+ ],
+ [
+ "Plop",
+ 0.085
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.085,
+ "floor_db": -46.5,
+ "stereo_corr": 1.0,
+ "darken": -0.53,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -82.1,
+ "first_sample": 0.0,
+ "tail_db": -55.6,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/bench/fs-545201_1.wav",
+ "length_s": 0.169,
+ "typ_z": 2.92,
+ "nn_cos": 0.816,
+ "tone_db": 13.86,
+ "tone_ratio": 1.03,
+ "n_outside": 7,
+ "outside": [
+ [
+ "1.26 kHz",
+ -32.8,
+ "below"
+ ],
+ [
+ "3.17 kHz",
+ -29.3,
+ "below"
+ ],
+ [
+ "4 kHz",
+ -24.2,
+ "below"
+ ]
+ ],
+ "attack_ms": 2.0,
+ "t40_ms": 125.0,
+ "active_s": 0.145,
+ "onset_s": 0.0,
+ "env_oct": 0.179,
+ "curve_db": 8.51,
+ "t40_ratio": 1.25,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.16,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 100.0,
+ "active_s": 0.125
+ },
+ "phrase_z": 2.8,
+ "phrase_min": 2.54,
+ "margin": 0.17,
+ "wins": 0.5,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Plop",
+ 0.425
+ ],
+ [
+ "Sound effect",
+ 0.102
+ ],
+ [
+ "Speech",
+ 0.078
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.425,
+ "floor_db": -47.9,
+ "stereo_corr": -0.666,
+ "darken": 0.74,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -80.8,
+ "first_sample": 0.0,
+ "tail_db": -60.8,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/bench/fs-573152_1.wav",
+ "length_s": 0.08,
+ "typ_z": 3.55,
+ "nn_cos": 0.832,
+ "tone_db": 11.03,
+ "tone_ratio": 0.8,
+ "n_outside": 3,
+ "outside": [
+ [
+ "5.04 kHz",
+ -23.8,
+ "below"
+ ],
+ [
+ "16 kHz",
+ -11.1,
+ "below"
+ ],
+ [
+ "12.7 kHz",
+ -10.4,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 70.0,
+ "active_s": 0.07,
+ "onset_s": 0.0,
+ "env_oct": 0.905,
+ "curve_db": 16.32,
+ "t40_ratio": 0.583,
+ "attack_ratio": 0.5,
+ "active_ratio": 0.522,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 120.0,
+ "active_s": 0.134
+ },
+ "phrase_z": 2.5,
+ "phrase_min": 2.47,
+ "margin": -0.98,
+ "wins": 0.0,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Speech",
+ 0.177
+ ],
+ [
+ "Throat clearing",
+ 0.057
+ ],
+ [
+ "Music",
+ 0.053
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.003,
+ "floor_db": -37.9,
+ "stereo_corr": 1.0,
+ "darken": 0.3,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -56.0,
+ "first_sample": 0.0,
+ "tail_db": -43.0,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/bench/mk-3192_1.wav",
+ "length_s": 0.215,
+ "typ_z": -0.78,
+ "nn_cos": 0.568,
+ "tone_db": 18.28,
+ "tone_ratio": 1.36,
+ "n_outside": 15,
+ "outside": [
+ [
+ "5.04 kHz",
+ 34.4,
+ "above"
+ ],
+ [
+ "6.35 kHz",
+ 31.5,
+ "above"
+ ],
+ [
+ "630 Hz",
+ -28.2,
+ "below"
+ ]
+ ],
+ "attack_ms": 2.0,
+ "t40_ms": 65.0,
+ "active_s": 0.134,
+ "onset_s": 0.016,
+ "env_oct": 0.328,
+ "curve_db": 11.88,
+ "t40_ratio": 0.542,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.072,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 120.0,
+ "active_s": 0.125
+ },
+ "phrase_z": 1.41,
+ "phrase_min": 0.97,
+ "margin": -1.83,
+ "wins": 0.0,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Speech",
+ 0.05
+ ],
+ [
+ "Plop",
+ 0.05
+ ],
+ [
+ "Coin (dropping)",
+ 0.044
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.05,
+ "floor_db": -52.5,
+ "stereo_corr": 1.0,
+ "darken": 1.71,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -73.3,
+ "first_sample": 0.0,
+ "tail_db": -69.2,
+ "lead_s": 0.006,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/bench/mk-3070_1.wav",
+ "length_s": 0.145,
+ "typ_z": -1.33,
+ "nn_cos": 0.642,
+ "tone_db": 12.85,
+ "tone_ratio": 0.93,
+ "n_outside": 3,
+ "outside": [
+ [
+ "2.52 kHz",
+ 22.7,
+ "above"
+ ],
+ [
+ "2 kHz",
+ 19.2,
+ "above"
+ ],
+ [
+ "315 Hz",
+ 16.5,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 120.0,
+ "active_s": 0.118,
+ "onset_s": 0.002,
+ "env_oct": 0.482,
+ "curve_db": 9.78,
+ "t40_ratio": 1.2,
+ "attack_ratio": 0.5,
+ "active_ratio": 0.881,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 100.0,
+ "active_s": 0.134
+ },
+ "phrase_z": 1.21,
+ "phrase_min": 0.86,
+ "margin": -1.6,
+ "wins": 0.0,
+ "best_confusion": "a hand clap",
+ "as_top": [
+ [
+ "Gasp",
+ 0.135
+ ],
+ [
+ "Speech",
+ 0.039
+ ],
+ [
+ "Scrape",
+ 0.037
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.006,
+ "floor_db": -56.7,
+ "stereo_corr": 0.87,
+ "darken": -0.05,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -76.2,
+ "first_sample": 0.0,
+ "tail_db": -77.0,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_rev": [
+ {
+ "file": "E:/earshot/out/refs/pop/bench/fs-202230_1.wav",
+ "length_s": 0.161,
+ "typ_z": -1.02,
+ "nn_cos": 0.559,
+ "tone_db": 13.87,
+ "tone_ratio": 0.89,
+ "n_outside": 9,
+ "outside": [
+ [
+ "157 Hz",
+ -29.4,
+ "below"
+ ],
+ [
+ "125 Hz",
+ -28.6,
+ "below"
+ ],
+ [
+ "99 Hz",
+ -22.9,
+ "below"
+ ]
+ ],
+ "attack_ms": 2.0,
+ "t40_ms": 140.0,
+ "active_s": 0.139,
+ "onset_s": 0.006,
+ "env_oct": 0.765,
+ "curve_db": 6.72,
+ "t40_ratio": 1.4,
+ "attack_ratio": 0.333,
+ "active_ratio": 1.168,
+ "ref_median": {
+ "attack_ms": 6.0,
+ "t40_ms": 100.0,
+ "active_s": 0.119
+ },
+ "phrase_z": 1.16,
+ "phrase_min": 1.03,
+ "margin": -1.14,
+ "wins": 0.0,
+ "best_confusion": "a hand clap",
+ "as_top": [
+ [
+ "Plop",
+ 0.159
+ ],
+ [
+ "Boing",
+ 0.131
+ ],
+ [
+ "Frog",
+ 0.102
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.159,
+ "floor_db": -41.9,
+ "stereo_corr": 1.0,
+ "darken": -0.25,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -116.1,
+ "first_sample": 0.0,
+ "tail_db": -48.0,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/bench/fs-33369_1.wav",
+ "length_s": 0.159,
+ "typ_z": 1.53,
+ "nn_cos": 0.933,
+ "tone_db": 12.12,
+ "tone_ratio": 0.78,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 5.0,
+ "t40_ms": 100.0,
+ "active_s": 0.125,
+ "onset_s": 0.005,
+ "env_oct": 0.111,
+ "curve_db": 3.86,
+ "t40_ratio": 1.0,
+ "attack_ratio": 0.833,
+ "active_ratio": 1.05,
+ "ref_median": {
+ "attack_ms": 6.0,
+ "t40_ms": 100.0,
+ "active_s": 0.119
+ },
+ "phrase_z": 2.89,
+ "phrase_min": 2.72,
+ "margin": -0.46,
+ "wins": 0.0,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.141
+ ],
+ [
+ "Speech",
+ 0.094
+ ],
+ [
+ "Plop",
+ 0.085
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.085,
+ "floor_db": -46.5,
+ "stereo_corr": 1.0,
+ "darken": -0.53,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -82.1,
+ "first_sample": 0.0,
+ "tail_db": -55.6,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/bench/fs-545201_1.wav",
+ "length_s": 0.169,
+ "typ_z": 1.02,
+ "nn_cos": 0.81,
+ "tone_db": 15.15,
+ "tone_ratio": 0.97,
+ "n_outside": 3,
+ "outside": [
+ [
+ "3.17 kHz",
+ -37.4,
+ "below"
+ ],
+ [
+ "2.52 kHz",
+ -32.8,
+ "below"
+ ],
+ [
+ "397 Hz",
+ -6.5,
+ "below"
+ ]
+ ],
+ "attack_ms": 2.0,
+ "t40_ms": 125.0,
+ "active_s": 0.145,
+ "onset_s": 0.0,
+ "env_oct": 0.731,
+ "curve_db": 7.07,
+ "t40_ratio": 1.25,
+ "attack_ratio": 0.333,
+ "active_ratio": 1.218,
+ "ref_median": {
+ "attack_ms": 6.0,
+ "t40_ms": 100.0,
+ "active_s": 0.119
+ },
+ "phrase_z": 2.8,
+ "phrase_min": 2.54,
+ "margin": 0.17,
+ "wins": 0.5,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Plop",
+ 0.425
+ ],
+ [
+ "Sound effect",
+ 0.102
+ ],
+ [
+ "Speech",
+ 0.078
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.425,
+ "floor_db": -47.9,
+ "stereo_corr": -0.666,
+ "darken": 0.74,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -80.8,
+ "first_sample": 0.0,
+ "tail_db": -60.8,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/bench/fs-573152_1.wav",
+ "length_s": 0.08,
+ "typ_z": 0.98,
+ "nn_cos": 0.789,
+ "tone_db": 13.94,
+ "tone_ratio": 0.9,
+ "n_outside": 5,
+ "outside": [
+ [
+ "1.26 kHz",
+ 20.4,
+ "above"
+ ],
+ [
+ "5.04 kHz",
+ -16.8,
+ "below"
+ ],
+ [
+ "397 Hz",
+ -11.3,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 70.0,
+ "active_s": 0.07,
+ "onset_s": 0.0,
+ "env_oct": 1.288,
+ "curve_db": 9.79,
+ "t40_ratio": 0.7,
+ "attack_ratio": 0.167,
+ "active_ratio": 0.588,
+ "ref_median": {
+ "attack_ms": 6.0,
+ "t40_ms": 100.0,
+ "active_s": 0.119
+ },
+ "phrase_z": 2.5,
+ "phrase_min": 2.47,
+ "margin": -0.98,
+ "wins": 0.0,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Speech",
+ 0.177
+ ],
+ [
+ "Throat clearing",
+ 0.057
+ ],
+ [
+ "Music",
+ 0.053
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.003,
+ "floor_db": -37.9,
+ "stereo_corr": 1.0,
+ "darken": 0.3,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -56.0,
+ "first_sample": 0.0,
+ "tail_db": -43.0,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/bench/mk-3192_1.wav",
+ "length_s": 0.215,
+ "typ_z": -0.49,
+ "nn_cos": 0.706,
+ "tone_db": 16.77,
+ "tone_ratio": 1.08,
+ "n_outside": 11,
+ "outside": [
+ [
+ "315 Hz",
+ -24.7,
+ "below"
+ ],
+ [
+ "630 Hz",
+ -22.9,
+ "below"
+ ],
+ [
+ "250 Hz",
+ -22.8,
+ "below"
+ ]
+ ],
+ "attack_ms": 2.0,
+ "t40_ms": 65.0,
+ "active_s": 0.134,
+ "onset_s": 0.016,
+ "env_oct": 0.793,
+ "curve_db": 7.79,
+ "t40_ratio": 0.65,
+ "attack_ratio": 0.333,
+ "active_ratio": 1.126,
+ "ref_median": {
+ "attack_ms": 6.0,
+ "t40_ms": 100.0,
+ "active_s": 0.119
+ },
+ "phrase_z": 1.41,
+ "phrase_min": 0.97,
+ "margin": -1.83,
+ "wins": 0.0,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Speech",
+ 0.05
+ ],
+ [
+ "Plop",
+ 0.05
+ ],
+ [
+ "Coin (dropping)",
+ 0.044
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.05,
+ "floor_db": -52.5,
+ "stereo_corr": 1.0,
+ "darken": 1.71,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -73.3,
+ "first_sample": 0.0,
+ "tail_db": -69.2,
+ "lead_s": 0.006,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/bench/mk-3070_1.wav",
+ "length_s": 0.145,
+ "typ_z": -0.8,
+ "nn_cos": 0.604,
+ "tone_db": 7.28,
+ "tone_ratio": 0.47,
+ "n_outside": 1,
+ "outside": [
+ [
+ "1.26 kHz",
+ 14.9,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 120.0,
+ "active_s": 0.118,
+ "onset_s": 0.002,
+ "env_oct": 0.953,
+ "curve_db": 8.3,
+ "t40_ratio": 1.2,
+ "attack_ratio": 0.167,
+ "active_ratio": 0.992,
+ "ref_median": {
+ "attack_ms": 6.0,
+ "t40_ms": 100.0,
+ "active_s": 0.119
+ },
+ "phrase_z": 1.21,
+ "phrase_min": 0.86,
+ "margin": -1.6,
+ "wins": 0.0,
+ "best_confusion": "a hand clap",
+ "as_top": [
+ [
+ "Gasp",
+ 0.135
+ ],
+ [
+ "Speech",
+ 0.039
+ ],
+ [
+ "Scrape",
+ 0.037
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.006,
+ "floor_db": -56.7,
+ "stereo_corr": 0.87,
+ "darken": -0.05,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -76.2,
+ "first_sample": 0.0,
+ "tail_db": -77.0,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity_bench": {
+ "n": 6,
+ "cos_mean": 0.522,
+ "cos_max": 0.832,
+ "tone_min_db": 7.62,
+ "tone_mean_db": 16.34,
+ "near_duplicates": []
+ },
+ "diversity_forge": {
+ "n": 7,
+ "cos_mean": 0.501,
+ "cos_max": 0.978,
+ "tone_min_db": 3.32,
+ "tone_mean_db": 18.73,
+ "near_duplicates": []
+ },
+ "summary": {
+ "made": {},
+ "real_cross": {
+ "n": 7,
+ "typ_z": -0.09,
+ "tone_db": 16.06,
+ "env_oct": 1.51,
+ "curve_db": 20.54,
+ "phrase_z": 2.46,
+ "margin": -0.5,
+ "as_p": 0.0,
+ "tone_ratio": 1.18,
+ "t40_ratio": 0.91,
+ "wins": 0.0,
+ "floor_db": -53.2,
+ "stereo_corr": 1.0,
+ "darken": 0.21,
+ "as_names": 0.29,
+ "artifacts": 0,
+ "var_cos": 0.501
+ },
+ "real_loo": {
+ "n": 6,
+ "typ_z": 1.13,
+ "tone_db": 13.66,
+ "env_oct": 0.41,
+ "curve_db": 12.5,
+ "phrase_z": 1.96,
+ "margin": -1.06,
+ "as_p": 0.07,
+ "tone_ratio": 1.0,
+ "t40_ratio": 1.02,
+ "wins": 0.0,
+ "floor_db": -47.2,
+ "stereo_corr": 1.0,
+ "darken": 0.12,
+ "as_names": 0.67,
+ "artifacts": 0,
+ "var_cos": 0.522
+ }
+ }
+ },
+ "riser": {
+ "n_bench": 2,
+ "n_forge": 1,
+ "note": "thin: one designed trailer riser (forge); the bench is two real cymbal swells, no synth risers.",
+ "bench_refs": [
+ "E:/earshot/out/refs/riser/bench/kit-riser_02.wav",
+ "E:/earshot/out/refs/riser/bench/fs-425434_1.wav"
+ ],
+ "made": [],
+ "diversity": {
+ "n": 0
+ },
+ "real_cross": [
+ {
+ "file": "E:/earshot/out/refs/riser/forge/kit-riser_01.wav",
+ "length_s": 2.575,
+ "nn_cos": 0.391,
+ "tone_db": 17.61,
+ "tone_ratio": null,
+ "n_outside": 18,
+ "outside": [
+ [
+ "50 Hz",
+ 42.7,
+ "above"
+ ],
+ [
+ "39 Hz",
+ 35.4,
+ "above"
+ ],
+ [
+ "62 Hz",
+ 33.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 1261.0,
+ "t40_ms": 85.0,
+ "active_s": 1.303,
+ "onset_s": 1.272,
+ "env_oct": 1.006,
+ "curve_db": 20.04,
+ "t40_ratio": 0.5,
+ "attack_ratio": 0.505,
+ "active_ratio": 0.489,
+ "ref_median": {
+ "attack_ms": 2497.0,
+ "t40_ms": 170.0,
+ "active_s": 2.664
+ },
+ "phrase_z": 3.52,
+ "phrase_min": 2.4,
+ "margin": 1.06,
+ "wins": 0.5,
+ "best_confusion": "a jet engine starting",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.35
+ ],
+ [
+ "Explosion",
+ 0.143
+ ],
+ [
+ "Sound effect",
+ 0.143
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.35,
+ "floor_db": -49.9,
+ "stereo_corr": 0.002,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -53.6,
+ "first_sample": 0.0,
+ "tail_db": -12.0,
+ "lead_s": 0.272,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_loo": [
+ {
+ "file": "E:/earshot/out/refs/riser/bench/kit-riser_02.wav",
+ "length_s": 4.15,
+ "nn_cos": 0.767,
+ "tone_db": 13.22,
+ "tone_ratio": null,
+ "n_outside": 24,
+ "outside": [
+ [
+ "10.1 kHz",
+ -26.8,
+ "below"
+ ],
+ [
+ "12.7 kHz",
+ -24.2,
+ "below"
+ ],
+ [
+ "8 kHz",
+ -22.5,
+ "below"
+ ]
+ ],
+ "attack_ms": 2501.0,
+ "t40_ms": 45.0,
+ "active_s": 2.542,
+ "onset_s": 1.608,
+ "env_oct": 0.95,
+ "curve_db": 36.07,
+ "t40_ratio": 0.153,
+ "attack_ratio": 1.003,
+ "active_ratio": 0.912,
+ "ref_median": {
+ "attack_ms": 2493.0,
+ "t40_ms": 295.0,
+ "active_s": 2.787
+ },
+ "phrase_z": 2.85,
+ "phrase_min": 2.75,
+ "margin": 1.81,
+ "wins": 1.0,
+ "best_confusion": "a jet engine starting",
+ "as_top": [
+ [
+ "Music",
+ 0.5
+ ],
+ [
+ "Sonar",
+ 0.397
+ ],
+ [
+ "Gong",
+ 0.159
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.011,
+ "floor_db": -37.6,
+ "stereo_corr": 1.0,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -125.2,
+ "first_sample": 0.0,
+ "tail_db": -3.3,
+ "lead_s": 0.057,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/riser/bench/fs-425434_1.wav",
+ "length_s": 2.975,
+ "nn_cos": 0.767,
+ "tone_db": 13.22,
+ "tone_ratio": null,
+ "n_outside": 24,
+ "outside": [
+ [
+ "10.1 kHz",
+ 26.8,
+ "above"
+ ],
+ [
+ "12.7 kHz",
+ 24.2,
+ "above"
+ ],
+ [
+ "8 kHz",
+ 22.5,
+ "above"
+ ]
+ ],
+ "attack_ms": 2493.0,
+ "t40_ms": 295.0,
+ "active_s": 2.787,
+ "onset_s": 0.183,
+ "env_oct": 0.95,
+ "curve_db": 36.07,
+ "t40_ratio": 6.556,
+ "attack_ratio": 0.997,
+ "active_ratio": 1.096,
+ "ref_median": {
+ "attack_ms": 2501.0,
+ "t40_ms": 45.0,
+ "active_s": 2.542
+ },
+ "phrase_z": 2.55,
+ "phrase_min": 2.29,
+ "margin": 0.49,
+ "wins": 0.5,
+ "best_confusion": "a jet engine starting",
+ "as_top": [
+ [
+ "Music",
+ 0.521
+ ],
+ [
+ "Sound effect",
+ 0.125
+ ],
+ [
+ "Whoosh, swoosh, swish",
+ 0.108
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.108,
+ "floor_db": -23.7,
+ "stereo_corr": 1.0,
+ "darken": 0.12,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -91.0,
+ "first_sample": 0.0,
+ "tail_db": -40.4,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_rev": [
+ {
+ "file": "E:/earshot/out/refs/riser/bench/kit-riser_02.wav",
+ "length_s": 4.15,
+ "nn_cos": 0.39,
+ "tone_db": 21.66,
+ "tone_ratio": null,
+ "n_outside": 21,
+ "outside": [
+ [
+ "50 Hz",
+ -51.2,
+ "below"
+ ],
+ [
+ "39 Hz",
+ -46.4,
+ "below"
+ ],
+ [
+ "62 Hz",
+ -44.2,
+ "below"
+ ]
+ ],
+ "attack_ms": 2501.0,
+ "t40_ms": 45.0,
+ "active_s": 2.542,
+ "onset_s": 1.608,
+ "env_oct": 0.957,
+ "curve_db": 13.25,
+ "t40_ratio": 0.529,
+ "attack_ratio": 1.983,
+ "active_ratio": 1.951,
+ "ref_median": {
+ "attack_ms": 1261.0,
+ "t40_ms": 85.0,
+ "active_s": 1.303
+ },
+ "phrase_z": 2.85,
+ "phrase_min": 2.75,
+ "margin": 1.81,
+ "wins": 1.0,
+ "best_confusion": "a jet engine starting",
+ "as_top": [
+ [
+ "Music",
+ 0.5
+ ],
+ [
+ "Sonar",
+ 0.397
+ ],
+ [
+ "Gong",
+ 0.159
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.011,
+ "floor_db": -37.6,
+ "stereo_corr": 1.0,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -125.2,
+ "first_sample": 0.0,
+ "tail_db": -3.3,
+ "lead_s": 0.057,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/riser/bench/fs-425434_1.wav",
+ "length_s": 2.975,
+ "nn_cos": 0.391,
+ "tone_db": 15.43,
+ "tone_ratio": null,
+ "n_outside": 26,
+ "outside": [
+ [
+ "50 Hz",
+ -34.3,
+ "below"
+ ],
+ [
+ "39 Hz",
+ -24.5,
+ "below"
+ ],
+ [
+ "79 Hz",
+ -23.6,
+ "below"
+ ]
+ ],
+ "attack_ms": 2493.0,
+ "t40_ms": 295.0,
+ "active_s": 2.787,
+ "onset_s": 0.183,
+ "env_oct": 1.292,
+ "curve_db": 31.87,
+ "t40_ratio": 3.471,
+ "attack_ratio": 1.977,
+ "active_ratio": 2.139,
+ "ref_median": {
+ "attack_ms": 1261.0,
+ "t40_ms": 85.0,
+ "active_s": 1.303
+ },
+ "phrase_z": 2.55,
+ "phrase_min": 2.29,
+ "margin": 0.49,
+ "wins": 0.5,
+ "best_confusion": "a jet engine starting",
+ "as_top": [
+ [
+ "Music",
+ 0.521
+ ],
+ [
+ "Sound effect",
+ 0.125
+ ],
+ [
+ "Whoosh, swoosh, swish",
+ 0.108
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.108,
+ "floor_db": -23.7,
+ "stereo_corr": 1.0,
+ "darken": 0.12,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -91.0,
+ "first_sample": 0.0,
+ "tail_db": -40.4,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity_bench": {
+ "n": 2,
+ "cos_mean": 0.767,
+ "cos_max": 0.767,
+ "tone_min_db": 13.22,
+ "tone_mean_db": 13.22,
+ "near_duplicates": []
+ },
+ "diversity_forge": {
+ "n": 1
+ },
+ "summary": {
+ "made": {},
+ "real_cross": {
+ "n": 1,
+ "tone_db": 17.61,
+ "env_oct": 1.01,
+ "curve_db": 20.04,
+ "phrase_z": 3.52,
+ "margin": 1.06,
+ "as_p": 0.35,
+ "t40_ratio": 0.5,
+ "wins": 0.5,
+ "floor_db": -49.9,
+ "stereo_corr": 0.0,
+ "as_names": 1.0,
+ "artifacts": 0
+ },
+ "real_loo": {
+ "n": 2,
+ "tone_db": 13.22,
+ "env_oct": 0.95,
+ "curve_db": 36.07,
+ "phrase_z": 2.7,
+ "margin": 1.15,
+ "as_p": 0.06,
+ "t40_ratio": 3.35,
+ "wins": 0.75,
+ "floor_db": -30.65,
+ "stereo_corr": 1.0,
+ "darken": 0.12,
+ "as_names": 0.5,
+ "artifacts": 0,
+ "var_cos": 0.767
+ }
+ }
+ },
+ "shimmer": {
+ "n_bench": 4,
+ "n_forge": 5,
+ "note": "kit shimmer_01 (bar chimes sweep) is bench: its author also recorded the celesta notify bench.",
+ "bench_refs": [
+ "E:/earshot/out/refs/shimmer/bench/kit-shimmer_01.wav",
+ "E:/earshot/out/refs/shimmer/bench/fs-257558_1.wav",
+ "E:/earshot/out/refs/shimmer/bench/vsco-BellTree_Stroke1_1.wav",
+ "E:/earshot/out/refs/shimmer/bench/vsco-glock_fx_up_chromatic_slow_01_1.wav"
+ ],
+ "made": [],
+ "diversity": {
+ "n": 0
+ },
+ "real_cross": [
+ {
+ "file": "E:/earshot/out/refs/shimmer/forge/kit-shimmer_02.wav",
+ "length_s": 2.34,
+ "typ_z": 1.69,
+ "nn_cos": 0.758,
+ "tone_db": 14.67,
+ "tone_ratio": 0.95,
+ "n_outside": 9,
+ "outside": [
+ [
+ "1.26 kHz",
+ 26.6,
+ "above"
+ ],
+ [
+ "2 kHz",
+ 18.6,
+ "above"
+ ],
+ [
+ "630 Hz",
+ 18.2,
+ "above"
+ ]
+ ],
+ "attack_ms": 200.0,
+ "t40_ms": 1720.0,
+ "active_s": 1.937,
+ "onset_s": 0.038,
+ "env_oct": 0.871,
+ "curve_db": 8.58,
+ "t40_ratio": 0.711,
+ "attack_ratio": 0.494,
+ "active_ratio": 0.465,
+ "ref_median": {
+ "attack_ms": 404.5,
+ "t40_ms": 2420.0,
+ "active_s": 4.168
+ },
+ "phrase_z": 2.87,
+ "phrase_min": 2.67,
+ "margin": 0.03,
+ "wins": 0.5,
+ "best_confusion": "a sleigh bell jingle",
+ "as_top": [
+ [
+ "Ding",
+ 0.532
+ ],
+ [
+ "Ping",
+ 0.057
+ ],
+ [
+ "Sound effect",
+ 0.034
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.02,
+ "floor_db": -46.8,
+ "stereo_corr": 0.712,
+ "darken": -1.07,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -87.4,
+ "first_sample": 0.0,
+ "tail_db": -96.6,
+ "lead_s": 0.015,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/shimmer/forge/fs-271355_1.wav",
+ "length_s": 2.623,
+ "typ_z": 0.81,
+ "nn_cos": 0.762,
+ "tone_db": 9.66,
+ "tone_ratio": 0.62,
+ "n_outside": 2,
+ "outside": [
+ [
+ "16 kHz",
+ 27.3,
+ "above"
+ ],
+ [
+ "12.7 kHz",
+ 15.4,
+ "above"
+ ]
+ ],
+ "attack_ms": 158.0,
+ "t40_ms": 2465.0,
+ "active_s": 2.62,
+ "onset_s": 0.0,
+ "env_oct": 0.684,
+ "curve_db": 7.75,
+ "t40_ratio": 1.019,
+ "attack_ratio": 0.391,
+ "active_ratio": 0.629,
+ "ref_median": {
+ "attack_ms": 404.5,
+ "t40_ms": 2420.0,
+ "active_s": 4.168
+ },
+ "phrase_z": 2.85,
+ "phrase_min": 2.78,
+ "margin": -0.06,
+ "wins": 0.5,
+ "best_confusion": "a sleigh bell jingle",
+ "as_top": [
+ [
+ "Wind chime",
+ 0.3
+ ],
+ [
+ "Chime",
+ 0.171
+ ],
+ [
+ "Music",
+ 0.06
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.3,
+ "floor_db": -21.1,
+ "stereo_corr": 0.171,
+ "darken": -0.19,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -67.9,
+ "first_sample": 0.0,
+ "tail_db": -38.0,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/shimmer/forge/mk-2014_1.wav",
+ "length_s": 2.215,
+ "typ_z": -3.72,
+ "nn_cos": 0.537,
+ "tone_db": 25.06,
+ "tone_ratio": 1.62,
+ "n_outside": 13,
+ "outside": [
+ [
+ "500 Hz",
+ 51.9,
+ "above"
+ ],
+ [
+ "397 Hz",
+ 49.3,
+ "above"
+ ],
+ [
+ "794 Hz",
+ 47.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 250.0,
+ "t40_ms": 1770.0,
+ "active_s": 2.02,
+ "onset_s": 0.0,
+ "env_oct": 0.73,
+ "curve_db": 10.59,
+ "t40_ratio": 0.731,
+ "attack_ratio": 0.618,
+ "active_ratio": 0.485,
+ "ref_median": {
+ "attack_ms": 404.5,
+ "t40_ms": 2420.0,
+ "active_s": 4.168
+ },
+ "phrase_z": 1.51,
+ "phrase_min": 1.41,
+ "margin": -0.28,
+ "wins": 0.0,
+ "best_confusion": "a sleigh bell jingle",
+ "as_top": [
+ [
+ "Music",
+ 0.293
+ ],
+ [
+ "Ringtone",
+ 0.111
+ ],
+ [
+ "Ding-dong",
+ 0.089
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.01,
+ "floor_db": -39.5,
+ "stereo_corr": 1.0,
+ "darken": -0.29,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -47.6,
+ "first_sample": 0.0,
+ "tail_db": -58.9,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/shimmer/forge/fs-495327_1.wav",
+ "length_s": 5.0,
+ "typ_z": 3.08,
+ "nn_cos": 0.862,
+ "tone_db": 7.21,
+ "tone_ratio": 0.46,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 1373.0,
+ "t40_ms": 375.0,
+ "active_s": 1.466,
+ "onset_s": 3.534,
+ "env_oct": 1.987,
+ "curve_db": 21.71,
+ "t40_ratio": 0.155,
+ "attack_ratio": 3.394,
+ "active_ratio": 0.352,
+ "ref_median": {
+ "attack_ms": 404.5,
+ "t40_ms": 2420.0,
+ "active_s": 4.168
+ },
+ "phrase_z": 3.76,
+ "phrase_min": 3.37,
+ "margin": 0.84,
+ "wins": 1.0,
+ "best_confusion": "a music box",
+ "as_top": [
+ [
+ "Wind chime",
+ 0.857
+ ],
+ [
+ "Chime",
+ 0.856
+ ],
+ [
+ "Music",
+ 0.095
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.857,
+ "floor_db": -120.0,
+ "stereo_corr": -0.003,
+ "darken": -0.21,
+ "artifacts": [
+ "tail cut"
+ ],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -116.6,
+ "first_sample": 0.0,
+ "tail_db": -22.4,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/shimmer/forge/fs-495327_2.wav",
+ "length_s": 5.0,
+ "typ_z": 2.56,
+ "nn_cos": 0.81,
+ "tone_db": 7.49,
+ "tone_ratio": 0.48,
+ "n_outside": 2,
+ "outside": [
+ [
+ "16 kHz",
+ 25.4,
+ "above"
+ ],
+ [
+ "12.7 kHz",
+ 13.0,
+ "above"
+ ]
+ ],
+ "attack_ms": 850.0,
+ "t40_ms": 2620.0,
+ "active_s": 3.468,
+ "onset_s": 1.527,
+ "env_oct": 0.484,
+ "curve_db": 8.13,
+ "t40_ratio": 1.083,
+ "attack_ratio": 2.101,
+ "active_ratio": 0.832,
+ "ref_median": {
+ "attack_ms": 404.5,
+ "t40_ms": 2420.0,
+ "active_s": 4.168
+ },
+ "phrase_z": 3.32,
+ "phrase_min": 2.93,
+ "margin": 0.48,
+ "wins": 1.0,
+ "best_confusion": "a music box",
+ "as_top": [
+ [
+ "Wind chime",
+ 0.792
+ ],
+ [
+ "Chime",
+ 0.693
+ ],
+ [
+ "Music",
+ 0.067
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.792,
+ "floor_db": -120.0,
+ "stereo_corr": -0.024,
+ "darken": -0.81,
+ "artifacts": [
+ "late start"
+ ],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -116.2,
+ "first_sample": 0.0,
+ "tail_db": -39.6,
+ "lead_s": 1.5,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_loo": [
+ {
+ "file": "E:/earshot/out/refs/shimmer/bench/kit-shimmer_01.wav",
+ "length_s": 6.0,
+ "typ_z": 5.2,
+ "nn_cos": 0.765,
+ "tone_db": 11.47,
+ "tone_ratio": 0.63,
+ "n_outside": 1,
+ "outside": [
+ [
+ "12.7 kHz",
+ 13.2,
+ "above"
+ ]
+ ],
+ "attack_ms": 2423.0,
+ "t40_ms": 3310.0,
+ "active_s": 5.742,
+ "onset_s": 0.173,
+ "env_oct": 1.72,
+ "curve_db": 12.46,
+ "t40_ratio": 1.619,
+ "attack_ratio": 10.489,
+ "active_ratio": 2.106,
+ "ref_median": {
+ "attack_ms": 231.0,
+ "t40_ms": 2045.0,
+ "active_s": 2.726
+ },
+ "phrase_z": 3.58,
+ "phrase_min": 3.5,
+ "margin": 0.58,
+ "wins": 1.0,
+ "best_confusion": "a sleigh bell jingle",
+ "as_top": [
+ [
+ "Chime",
+ 0.408
+ ],
+ [
+ "Wind chime",
+ 0.358
+ ],
+ [
+ "Sound effect",
+ 0.048
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.408,
+ "floor_db": -29.5,
+ "stereo_corr": -0.017,
+ "darken": -0.19,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -120.9,
+ "first_sample": 0.0,
+ "tail_db": -66.6,
+ "lead_s": 0.01,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/shimmer/bench/fs-257558_1.wav",
+ "length_s": 1.904,
+ "typ_z": 2.4,
+ "nn_cos": 0.765,
+ "tone_db": 18.35,
+ "tone_ratio": 1.45,
+ "n_outside": 11,
+ "outside": [
+ [
+ "16 kHz",
+ -43.0,
+ "below"
+ ],
+ [
+ "1.59 kHz",
+ -36.7,
+ "below"
+ ],
+ [
+ "1.26 kHz",
+ -30.5,
+ "below"
+ ]
+ ],
+ "attack_ms": 7.0,
+ "t40_ms": 1660.0,
+ "active_s": 1.893,
+ "onset_s": 0.002,
+ "env_oct": 2.896,
+ "curve_db": 13.58,
+ "t40_ratio": 0.594,
+ "attack_ratio": 0.012,
+ "active_ratio": 0.337,
+ "ref_median": {
+ "attack_ms": 578.0,
+ "t40_ms": 2795.0,
+ "active_s": 5.61
+ },
+ "phrase_z": 4.03,
+ "phrase_min": 3.82,
+ "margin": 0.55,
+ "wins": 1.0,
+ "best_confusion": "a sleigh bell jingle",
+ "as_top": [
+ [
+ "Music",
+ 0.386
+ ],
+ [
+ "Ringtone",
+ 0.278
+ ],
+ [
+ "Jingle, tinkle",
+ 0.177
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.177,
+ "floor_db": -26.1,
+ "stereo_corr": 0.921,
+ "darken": 0.15,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -145.6,
+ "first_sample": 0.0,
+ "tail_db": -42.8,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/shimmer/bench/vsco-BellTree_Stroke1_1.wav",
+ "length_s": 3.275,
+ "typ_z": -2.65,
+ "nn_cos": 0.679,
+ "tone_db": 12.68,
+ "tone_ratio": 0.69,
+ "n_outside": 4,
+ "outside": [
+ [
+ "16 kHz",
+ 35.7,
+ "above"
+ ],
+ [
+ "500 Hz",
+ 15.1,
+ "above"
+ ],
+ [
+ "397 Hz",
+ 14.2,
+ "above"
+ ]
+ ],
+ "attack_ms": 578.0,
+ "t40_ms": 2045.0,
+ "active_s": 2.726,
+ "onset_s": 0.154,
+ "env_oct": 0.938,
+ "curve_db": 9.83,
+ "t40_ratio": 0.732,
+ "attack_ratio": 2.502,
+ "active_ratio": 0.486,
+ "ref_median": {
+ "attack_ms": 231.0,
+ "t40_ms": 2795.0,
+ "active_s": 5.61
+ },
+ "phrase_z": 3.09,
+ "phrase_min": 2.87,
+ "margin": 0.1,
+ "wins": 0.5,
+ "best_confusion": "a sleigh bell jingle",
+ "as_top": [
+ [
+ "Music",
+ 0.199
+ ],
+ [
+ "Ringtone",
+ 0.129
+ ],
+ [
+ "Wind chime",
+ 0.121
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.121,
+ "floor_db": -42.7,
+ "stereo_corr": 0.431,
+ "darken": -0.48,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -82.0,
+ "first_sample": 0.0,
+ "tail_db": -64.6,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/shimmer/bench/vsco-glock_fx_up_chromatic_slow_01_1.wav",
+ "length_s": 6.0,
+ "typ_z": -0.37,
+ "nn_cos": 0.667,
+ "tone_db": 21.31,
+ "tone_ratio": 1.68,
+ "n_outside": 17,
+ "outside": [
+ [
+ "1 kHz",
+ 48.4,
+ "above"
+ ],
+ [
+ "794 Hz",
+ 40.6,
+ "above"
+ ],
+ [
+ "12.7 kHz",
+ -37.6,
+ "below"
+ ]
+ ],
+ "attack_ms": 231.0,
+ "t40_ms": 2795.0,
+ "active_s": 5.61,
+ "onset_s": 0.09,
+ "env_oct": 0.938,
+ "curve_db": 13.33,
+ "t40_ratio": 1.367,
+ "attack_ratio": 0.4,
+ "active_ratio": 2.058,
+ "ref_median": {
+ "attack_ms": 578.0,
+ "t40_ms": 2045.0,
+ "active_s": 2.726
+ },
+ "phrase_z": 3.22,
+ "phrase_min": 2.99,
+ "margin": 0.56,
+ "wins": 1.0,
+ "best_confusion": "a harp glissando",
+ "as_top": [
+ [
+ "Chime",
+ 0.231
+ ],
+ [
+ "Music",
+ 0.187
+ ],
+ [
+ "Glockenspiel",
+ 0.133
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.231,
+ "floor_db": -49.1,
+ "stereo_corr": 0.926,
+ "darken": -1.17,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -86.8,
+ "first_sample": 0.0,
+ "tail_db": -84.1,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_rev": [
+ {
+ "file": "E:/earshot/out/refs/shimmer/bench/kit-shimmer_01.wav",
+ "length_s": 6.0,
+ "typ_z": 1.45,
+ "nn_cos": 0.862,
+ "tone_db": 7.44,
+ "tone_ratio": 0.61,
+ "n_outside": 2,
+ "outside": [
+ [
+ "1 kHz",
+ -18.8,
+ "below"
+ ],
+ [
+ "794 Hz",
+ -18.5,
+ "below"
+ ]
+ ],
+ "attack_ms": 2423.0,
+ "t40_ms": 3310.0,
+ "active_s": 5.742,
+ "onset_s": 0.173,
+ "env_oct": 1.896,
+ "curve_db": 22.1,
+ "t40_ratio": 1.87,
+ "attack_ratio": 9.692,
+ "active_ratio": 2.843,
+ "ref_median": {
+ "attack_ms": 250.0,
+ "t40_ms": 1770.0,
+ "active_s": 2.02
+ },
+ "phrase_z": 3.58,
+ "phrase_min": 3.5,
+ "margin": 0.58,
+ "wins": 1.0,
+ "best_confusion": "a sleigh bell jingle",
+ "as_top": [
+ [
+ "Chime",
+ 0.408
+ ],
+ [
+ "Wind chime",
+ 0.358
+ ],
+ [
+ "Sound effect",
+ 0.048
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.408,
+ "floor_db": -29.5,
+ "stereo_corr": -0.017,
+ "darken": -0.19,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -120.9,
+ "first_sample": 0.0,
+ "tail_db": -66.6,
+ "lead_s": 0.01,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/shimmer/bench/fs-257558_1.wav",
+ "length_s": 1.904,
+ "typ_z": 0.46,
+ "nn_cos": 0.753,
+ "tone_db": 16.07,
+ "tone_ratio": 1.32,
+ "n_outside": 10,
+ "outside": [
+ [
+ "16 kHz",
+ -48.9,
+ "below"
+ ],
+ [
+ "1.59 kHz",
+ -30.6,
+ "below"
+ ],
+ [
+ "1.26 kHz",
+ -29.7,
+ "below"
+ ]
+ ],
+ "attack_ms": 7.0,
+ "t40_ms": 1660.0,
+ "active_s": 1.893,
+ "onset_s": 0.002,
+ "env_oct": 1.782,
+ "curve_db": 7.87,
+ "t40_ratio": 0.938,
+ "attack_ratio": 0.028,
+ "active_ratio": 0.937,
+ "ref_median": {
+ "attack_ms": 250.0,
+ "t40_ms": 1770.0,
+ "active_s": 2.02
+ },
+ "phrase_z": 4.03,
+ "phrase_min": 3.82,
+ "margin": 0.55,
+ "wins": 1.0,
+ "best_confusion": "a sleigh bell jingle",
+ "as_top": [
+ [
+ "Music",
+ 0.386
+ ],
+ [
+ "Ringtone",
+ 0.278
+ ],
+ [
+ "Jingle, tinkle",
+ 0.177
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.177,
+ "floor_db": -26.1,
+ "stereo_corr": 0.921,
+ "darken": 0.15,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -145.6,
+ "first_sample": 0.0,
+ "tail_db": -42.8,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/shimmer/bench/vsco-BellTree_Stroke1_1.wav",
+ "length_s": 3.275,
+ "typ_z": 0.55,
+ "nn_cos": 0.746,
+ "tone_db": 8.78,
+ "tone_ratio": 0.72,
+ "n_outside": 1,
+ "outside": [
+ [
+ "3.17 kHz",
+ 5.2,
+ "above"
+ ]
+ ],
+ "attack_ms": 578.0,
+ "t40_ms": 2045.0,
+ "active_s": 2.726,
+ "onset_s": 0.154,
+ "env_oct": 0.617,
+ "curve_db": 8.42,
+ "t40_ratio": 1.155,
+ "attack_ratio": 2.312,
+ "active_ratio": 1.35,
+ "ref_median": {
+ "attack_ms": 250.0,
+ "t40_ms": 1770.0,
+ "active_s": 2.02
+ },
+ "phrase_z": 3.09,
+ "phrase_min": 2.87,
+ "margin": 0.1,
+ "wins": 0.5,
+ "best_confusion": "a sleigh bell jingle",
+ "as_top": [
+ [
+ "Music",
+ 0.199
+ ],
+ [
+ "Ringtone",
+ 0.129
+ ],
+ [
+ "Wind chime",
+ 0.121
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.121,
+ "floor_db": -42.7,
+ "stereo_corr": 0.431,
+ "darken": -0.48,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -82.0,
+ "first_sample": 0.0,
+ "tail_db": -64.6,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/shimmer/bench/vsco-glock_fx_up_chromatic_slow_01_1.wav",
+ "length_s": 6.0,
+ "typ_z": 0.77,
+ "nn_cos": 0.758,
+ "tone_db": 18.91,
+ "tone_ratio": 1.56,
+ "n_outside": 6,
+ "outside": [
+ [
+ "12.7 kHz",
+ -44.2,
+ "below"
+ ],
+ [
+ "10.1 kHz",
+ -27.0,
+ "below"
+ ],
+ [
+ "8 kHz",
+ -14.2,
+ "below"
+ ]
+ ],
+ "attack_ms": 231.0,
+ "t40_ms": 2795.0,
+ "active_s": 5.61,
+ "onset_s": 0.09,
+ "env_oct": 0.749,
+ "curve_db": 17.14,
+ "t40_ratio": 1.579,
+ "attack_ratio": 0.924,
+ "active_ratio": 2.777,
+ "ref_median": {
+ "attack_ms": 250.0,
+ "t40_ms": 1770.0,
+ "active_s": 2.02
+ },
+ "phrase_z": 3.22,
+ "phrase_min": 2.99,
+ "margin": 0.56,
+ "wins": 1.0,
+ "best_confusion": "a harp glissando",
+ "as_top": [
+ [
+ "Chime",
+ 0.231
+ ],
+ [
+ "Music",
+ 0.187
+ ],
+ [
+ "Glockenspiel",
+ 0.133
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.231,
+ "floor_db": -49.1,
+ "stereo_corr": 0.926,
+ "darken": -1.17,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -86.8,
+ "first_sample": 0.0,
+ "tail_db": -84.1,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity_bench": {
+ "n": 4,
+ "cos_mean": 0.644,
+ "cos_max": 0.765,
+ "tone_min_db": 10.66,
+ "tone_mean_db": 17.82,
+ "near_duplicates": []
+ },
+ "diversity_forge": {
+ "n": 5,
+ "cos_mean": 0.59,
+ "cos_max": 0.955,
+ "tone_min_db": 2.67,
+ "tone_mean_db": 17.27,
+ "near_duplicates": []
+ },
+ "summary": {
+ "made": {},
+ "real_cross": {
+ "n": 5,
+ "typ_z": 1.69,
+ "tone_db": 9.66,
+ "env_oct": 0.73,
+ "curve_db": 8.58,
+ "phrase_z": 2.87,
+ "margin": 0.03,
+ "as_p": 0.3,
+ "tone_ratio": 0.62,
+ "t40_ratio": 0.73,
+ "wins": 0.5,
+ "floor_db": -46.8,
+ "stereo_corr": 0.17,
+ "darken": -0.29,
+ "as_names": 0.6,
+ "artifacts": 2,
+ "var_cos": 0.59
+ },
+ "real_loo": {
+ "n": 4,
+ "typ_z": 1.01,
+ "tone_db": 15.52,
+ "env_oct": 1.33,
+ "curve_db": 12.89,
+ "phrase_z": 3.4,
+ "margin": 0.56,
+ "as_p": 0.2,
+ "tone_ratio": 1.07,
+ "t40_ratio": 1.05,
+ "wins": 1.0,
+ "floor_db": -36.1,
+ "stereo_corr": 0.68,
+ "darken": -0.33,
+ "as_names": 1.0,
+ "artifacts": 0,
+ "var_cos": 0.644
+ }
+ }
+ },
+ "sub_drop": {
+ "n_bench": 1,
+ "n_forge": 0,
+ "note": "one designed source (a tom pitched down 24 semitones by its author), held out as bench; the forge has no sub-drop refs.",
+ "bench_refs": [
+ "E:/earshot/out/refs/sub_drop/bench/fs-428073_1.wav"
+ ],
+ "made": [],
+ "diversity": {
+ "n": 0
+ },
+ "real_cross": [],
+ "real_loo": [],
+ "real_rev": [],
+ "diversity_bench": {
+ "n": 1
+ },
+ "diversity_forge": {
+ "n": 0
+ },
+ "summary": {
+ "made": {},
+ "real_cross": {},
+ "real_loo": {}
+ }
+ },
+ "swipe": {
+ "n_bench": 4,
+ "n_forge": 4,
+ "note": "cloth only: the library has no paper slides.",
+ "bench_refs": [
+ "E:/earshot/out/refs/swipe/bench/kit-swipe_01.wav",
+ "E:/earshot/out/refs/swipe/bench/fs-495390_1.wav",
+ "E:/earshot/out/refs/swipe/bench/fs-495390_2.wav",
+ "E:/earshot/out/refs/swipe/bench/fs-495390_3.wav"
+ ],
+ "made": [],
+ "diversity": {
+ "n": 0
+ },
+ "real_cross": [
+ {
+ "file": "E:/earshot/out/refs/swipe/forge/kit-swipe_02.wav",
+ "length_s": 0.595,
+ "typ_z": -8.97,
+ "nn_cos": 0.662,
+ "tone_db": 16.38,
+ "tone_ratio": 4.5,
+ "n_outside": 19,
+ "outside": [
+ [
+ "12.7 kHz",
+ -34.5,
+ "below"
+ ],
+ [
+ "10.1 kHz",
+ -32.1,
+ "below"
+ ],
+ [
+ "16 kHz",
+ -29.2,
+ "below"
+ ]
+ ],
+ "attack_ms": 123.0,
+ "t40_ms": 145.0,
+ "active_s": 0.412,
+ "onset_s": 0.043,
+ "env_oct": 0.396,
+ "curve_db": 7.46,
+ "t40_ratio": 0.637,
+ "attack_ratio": 1.177,
+ "active_ratio": 1.234,
+ "ref_median": {
+ "attack_ms": 104.5,
+ "t40_ms": 227.5,
+ "active_s": 0.334
+ },
+ "phrase_z": 2.33,
+ "phrase_min": 2.17,
+ "margin": 0.83,
+ "wins": 0.0,
+ "best_confusion": "a broom sweeping",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.126
+ ],
+ [
+ "Sound effect",
+ 0.079
+ ],
+ [
+ "Bang",
+ 0.047
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.126,
+ "floor_db": -73.3,
+ "stereo_corr": 0.52,
+ "darken": 0.54,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -79.2,
+ "first_sample": 0.0,
+ "tail_db": -100.9,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/swipe/forge/kit-whoosh_med_03.wav",
+ "length_s": 0.665,
+ "typ_z": -11.17,
+ "nn_cos": 0.609,
+ "tone_db": 17.98,
+ "tone_ratio": 4.94,
+ "n_outside": 19,
+ "outside": [
+ [
+ "12.7 kHz",
+ -37.1,
+ "below"
+ ],
+ [
+ "10.1 kHz",
+ -35.5,
+ "below"
+ ],
+ [
+ "8 kHz",
+ -30.7,
+ "below"
+ ]
+ ],
+ "attack_ms": 78.0,
+ "t40_ms": 195.0,
+ "active_s": 0.477,
+ "onset_s": 0.063,
+ "env_oct": 0.386,
+ "curve_db": 14.63,
+ "t40_ratio": 0.857,
+ "attack_ratio": 0.746,
+ "active_ratio": 1.428,
+ "ref_median": {
+ "attack_ms": 104.5,
+ "t40_ms": 227.5,
+ "active_s": 0.334
+ },
+ "phrase_z": 2.71,
+ "phrase_min": 2.66,
+ "margin": 0.23,
+ "wins": 1.0,
+ "best_confusion": "a gust of wind",
+ "as_top": [
+ [
+ "Slam",
+ 0.133
+ ],
+ [
+ "Door",
+ 0.101
+ ],
+ [
+ "Whoosh, swoosh, swish",
+ 0.084
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.084,
+ "floor_db": -54.6,
+ "stereo_corr": -0.301,
+ "darken": -0.48,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -123.7,
+ "first_sample": 0.0,
+ "tail_db": -74.4,
+ "lead_s": 0.01,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/swipe/forge/fs-425856_1.wav",
+ "length_s": 0.848,
+ "typ_z": -11.64,
+ "nn_cos": 0.626,
+ "tone_db": 17.02,
+ "tone_ratio": 4.67,
+ "n_outside": 20,
+ "outside": [
+ [
+ "12.7 kHz",
+ -36.6,
+ "below"
+ ],
+ [
+ "10.1 kHz",
+ -34.5,
+ "below"
+ ],
+ [
+ "8 kHz",
+ -29.9,
+ "below"
+ ]
+ ],
+ "attack_ms": 49.0,
+ "t40_ms": 220.0,
+ "active_s": 0.404,
+ "onset_s": 0.126,
+ "env_oct": 0.472,
+ "curve_db": 16.64,
+ "t40_ratio": 0.967,
+ "attack_ratio": 0.469,
+ "active_ratio": 1.21,
+ "ref_median": {
+ "attack_ms": 104.5,
+ "t40_ms": 227.5,
+ "active_s": 0.334
+ },
+ "phrase_z": 2.69,
+ "phrase_min": 2.5,
+ "margin": 0.36,
+ "wins": 1.0,
+ "best_confusion": "a gust of wind",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.165
+ ],
+ [
+ "Sound effect",
+ 0.095
+ ],
+ [
+ "Slam",
+ 0.05
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.165,
+ "floor_db": -59.1,
+ "stereo_corr": 0.319,
+ "darken": 0.32,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -79.5,
+ "first_sample": 0.0,
+ "tail_db": -80.7,
+ "lead_s": 0.097,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/swipe/forge/fs-425856_2.wav",
+ "length_s": 0.623,
+ "typ_z": -7.87,
+ "nn_cos": 0.692,
+ "tone_db": 16.64,
+ "tone_ratio": 4.57,
+ "n_outside": 22,
+ "outside": [
+ [
+ "12.7 kHz",
+ -34.4,
+ "below"
+ ],
+ [
+ "10.1 kHz",
+ -31.4,
+ "below"
+ ],
+ [
+ "16 kHz",
+ -29.6,
+ "below"
+ ]
+ ],
+ "attack_ms": 134.0,
+ "t40_ms": 165.0,
+ "active_s": 0.535,
+ "onset_s": 0.03,
+ "env_oct": 0.501,
+ "curve_db": 13.61,
+ "t40_ratio": 0.725,
+ "attack_ratio": 1.282,
+ "active_ratio": 1.602,
+ "ref_median": {
+ "attack_ms": 104.5,
+ "t40_ms": 227.5,
+ "active_s": 0.334
+ },
+ "phrase_z": 3.38,
+ "phrase_min": 3.2,
+ "margin": 1.31,
+ "wins": 1.0,
+ "best_confusion": "a gust of wind",
+ "as_top": [
+ [
+ "Bang",
+ 0.118
+ ],
+ [
+ "Sound effect",
+ 0.113
+ ],
+ [
+ "Whoosh, swoosh, swish",
+ 0.085
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.085,
+ "floor_db": -44.6,
+ "stereo_corr": 0.611,
+ "darken": -0.19,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -76.4,
+ "first_sample": 0.0,
+ "tail_db": -69.9,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_loo": [
+ {
+ "file": "E:/earshot/out/refs/swipe/bench/kit-swipe_01.wav",
+ "length_s": 0.685,
+ "typ_z": 0.26,
+ "nn_cos": 0.874,
+ "tone_db": 3.17,
+ "tone_ratio": 0.77,
+ "n_outside": 2,
+ "outside": [
+ [
+ "1 kHz",
+ 5.2,
+ "above"
+ ],
+ [
+ "39 Hz",
+ -4.7,
+ "below"
+ ]
+ ],
+ "attack_ms": 100.0,
+ "t40_ms": 250.0,
+ "active_s": 0.344,
+ "onset_s": 0.151,
+ "env_oct": 0.157,
+ "curve_db": 9.3,
+ "t40_ratio": 1.22,
+ "attack_ratio": 0.935,
+ "active_ratio": 1.062,
+ "ref_median": {
+ "attack_ms": 107.0,
+ "t40_ms": 205.0,
+ "active_s": 0.324
+ },
+ "phrase_z": 3.64,
+ "phrase_min": 3.47,
+ "margin": 2.37,
+ "wins": 1.0,
+ "best_confusion": "a broom sweeping",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.167
+ ],
+ [
+ "Bang",
+ 0.11
+ ],
+ [
+ "Sound effect",
+ 0.061
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.167,
+ "floor_db": -64.0,
+ "stereo_corr": 0.869,
+ "darken": 0.52,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -78.4,
+ "first_sample": 0.0,
+ "tail_db": -87.4,
+ "lead_s": 0.03,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/swipe/bench/fs-495390_1.wav",
+ "length_s": 0.383,
+ "typ_z": -4.47,
+ "nn_cos": 0.885,
+ "tone_db": 4.4,
+ "tone_ratio": 1.39,
+ "n_outside": 4,
+ "outside": [
+ [
+ "31 Hz",
+ 10.1,
+ "above"
+ ],
+ [
+ "125 Hz",
+ -9.5,
+ "below"
+ ],
+ [
+ "500 Hz",
+ -5.5,
+ "below"
+ ]
+ ],
+ "attack_ms": 102.0,
+ "t40_ms": 205.0,
+ "active_s": 0.318,
+ "onset_s": 0.032,
+ "env_oct": 0.156,
+ "curve_db": 8.49,
+ "t40_ratio": 0.82,
+ "attack_ratio": 0.953,
+ "active_ratio": 0.924,
+ "ref_median": {
+ "attack_ms": 107.0,
+ "t40_ms": 250.0,
+ "active_s": 0.344
+ },
+ "phrase_z": 3.13,
+ "phrase_min": 3.01,
+ "margin": 1.66,
+ "wins": 1.0,
+ "best_confusion": "paper crumpling",
+ "as_top": [
+ [
+ "Bang",
+ 0.202
+ ],
+ [
+ "Sound effect",
+ 0.127
+ ],
+ [
+ "Burst, pop",
+ 0.036
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.019,
+ "floor_db": -38.7,
+ "stereo_corr": 0.667,
+ "darken": 0.71,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -79.3,
+ "first_sample": 0.0,
+ "tail_db": -63.5,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/swipe/bench/fs-495390_2.wav",
+ "length_s": 0.515,
+ "typ_z": 4.35,
+ "nn_cos": 0.918,
+ "tone_db": 3.17,
+ "tone_ratio": 0.77,
+ "n_outside": 1,
+ "outside": [
+ [
+ "62 Hz",
+ 7.2,
+ "above"
+ ]
+ ],
+ "attack_ms": 107.0,
+ "t40_ms": 290.0,
+ "active_s": 0.394,
+ "onset_s": 0.101,
+ "env_oct": 0.284,
+ "curve_db": 12.51,
+ "t40_ratio": 1.415,
+ "attack_ratio": 1.049,
+ "active_ratio": 1.216,
+ "ref_median": {
+ "attack_ms": 102.0,
+ "t40_ms": 205.0,
+ "active_s": 0.324
+ },
+ "phrase_z": 3.33,
+ "phrase_min": 3.16,
+ "margin": 1.91,
+ "wins": 1.0,
+ "best_confusion": "paper crumpling",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.071
+ ],
+ [
+ "Bang",
+ 0.065
+ ],
+ [
+ "Whip",
+ 0.061
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.045,
+ "floor_db": -35.2,
+ "stereo_corr": 0.255,
+ "darken": 0.14,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -69.7,
+ "first_sample": 0.0,
+ "tail_db": -51.4,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/swipe/bench/fs-495390_3.wav",
+ "length_s": 0.405,
+ "typ_z": 2.25,
+ "nn_cos": 0.918,
+ "tone_db": 4.11,
+ "tone_ratio": 1.3,
+ "n_outside": 3,
+ "outside": [
+ [
+ "125 Hz",
+ 11.2,
+ "above"
+ ],
+ [
+ "157 Hz",
+ 7.4,
+ "above"
+ ],
+ [
+ "16 kHz",
+ -6.5,
+ "below"
+ ]
+ ],
+ "attack_ms": 138.0,
+ "t40_ms": 185.0,
+ "active_s": 0.324,
+ "onset_s": 0.061,
+ "env_oct": 0.319,
+ "curve_db": 10.85,
+ "t40_ratio": 0.74,
+ "attack_ratio": 1.353,
+ "active_ratio": 0.942,
+ "ref_median": {
+ "attack_ms": 102.0,
+ "t40_ms": 250.0,
+ "active_s": 0.344
+ },
+ "phrase_z": 3.52,
+ "phrase_min": 3.4,
+ "margin": 2.24,
+ "wins": 1.0,
+ "best_confusion": "a zipper",
+ "as_top": [
+ [
+ "Bang",
+ 0.05
+ ],
+ [
+ "Sound effect",
+ 0.033
+ ],
+ [
+ "Burst, pop",
+ 0.024
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.021,
+ "floor_db": -37.4,
+ "stereo_corr": 0.698,
+ "darken": 0.81,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -72.8,
+ "first_sample": 0.0,
+ "tail_db": -50.7,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_rev": [
+ {
+ "file": "E:/earshot/out/refs/swipe/bench/kit-swipe_01.wav",
+ "length_s": 0.685,
+ "typ_z": -5.49,
+ "nn_cos": 0.692,
+ "tone_db": 17.91,
+ "tone_ratio": 6.54,
+ "n_outside": 20,
+ "outside": [
+ [
+ "12.7 kHz",
+ 36.8,
+ "above"
+ ],
+ [
+ "10.1 kHz",
+ 34.3,
+ "above"
+ ],
+ [
+ "16 kHz",
+ 31.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 100.0,
+ "t40_ms": 250.0,
+ "active_s": 0.344,
+ "onset_s": 0.151,
+ "env_oct": 0.284,
+ "curve_db": 8.34,
+ "t40_ratio": 1.389,
+ "attack_ratio": 0.995,
+ "active_ratio": 0.774,
+ "ref_median": {
+ "attack_ms": 100.5,
+ "t40_ms": 180.0,
+ "active_s": 0.445
+ },
+ "phrase_z": 3.64,
+ "phrase_min": 3.47,
+ "margin": 2.37,
+ "wins": 1.0,
+ "best_confusion": "a broom sweeping",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.167
+ ],
+ [
+ "Bang",
+ 0.11
+ ],
+ [
+ "Sound effect",
+ 0.061
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.167,
+ "floor_db": -64.0,
+ "stereo_corr": 0.869,
+ "darken": 0.52,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -78.4,
+ "first_sample": 0.0,
+ "tail_db": -87.4,
+ "lead_s": 0.03,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/swipe/bench/fs-495390_1.wav",
+ "length_s": 0.383,
+ "typ_z": -7.34,
+ "nn_cos": 0.635,
+ "tone_db": 15.72,
+ "tone_ratio": 5.74,
+ "n_outside": 19,
+ "outside": [
+ [
+ "12.7 kHz",
+ 34.3,
+ "above"
+ ],
+ [
+ "10.1 kHz",
+ 32.3,
+ "above"
+ ],
+ [
+ "16 kHz",
+ 28.3,
+ "above"
+ ]
+ ],
+ "attack_ms": 102.0,
+ "t40_ms": 205.0,
+ "active_s": 0.318,
+ "onset_s": 0.032,
+ "env_oct": 0.231,
+ "curve_db": 14.67,
+ "t40_ratio": 1.139,
+ "attack_ratio": 1.015,
+ "active_ratio": 0.715,
+ "ref_median": {
+ "attack_ms": 100.5,
+ "t40_ms": 180.0,
+ "active_s": 0.445
+ },
+ "phrase_z": 3.13,
+ "phrase_min": 3.01,
+ "margin": 1.66,
+ "wins": 1.0,
+ "best_confusion": "paper crumpling",
+ "as_top": [
+ [
+ "Bang",
+ 0.202
+ ],
+ [
+ "Sound effect",
+ 0.127
+ ],
+ [
+ "Burst, pop",
+ 0.036
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.019,
+ "floor_db": -38.7,
+ "stereo_corr": 0.667,
+ "darken": 0.71,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -79.3,
+ "first_sample": 0.0,
+ "tail_db": -63.5,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/swipe/bench/fs-495390_2.wav",
+ "length_s": 0.515,
+ "typ_z": -7.07,
+ "nn_cos": 0.636,
+ "tone_db": 17.64,
+ "tone_ratio": 6.44,
+ "n_outside": 18,
+ "outside": [
+ [
+ "12.7 kHz",
+ 36.9,
+ "above"
+ ],
+ [
+ "10.1 kHz",
+ 34.3,
+ "above"
+ ],
+ [
+ "16 kHz",
+ 31.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 107.0,
+ "t40_ms": 290.0,
+ "active_s": 0.394,
+ "onset_s": 0.101,
+ "env_oct": 0.317,
+ "curve_db": 10.3,
+ "t40_ratio": 1.611,
+ "attack_ratio": 1.065,
+ "active_ratio": 0.886,
+ "ref_median": {
+ "attack_ms": 100.5,
+ "t40_ms": 180.0,
+ "active_s": 0.445
+ },
+ "phrase_z": 3.33,
+ "phrase_min": 3.16,
+ "margin": 1.91,
+ "wins": 1.0,
+ "best_confusion": "paper crumpling",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.071
+ ],
+ [
+ "Bang",
+ 0.065
+ ],
+ [
+ "Whip",
+ 0.061
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.045,
+ "floor_db": -35.2,
+ "stereo_corr": 0.255,
+ "darken": 0.14,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -69.7,
+ "first_sample": 0.0,
+ "tail_db": -51.4,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/swipe/bench/fs-495390_3.wav",
+ "length_s": 0.405,
+ "typ_z": -5.7,
+ "nn_cos": 0.675,
+ "tone_db": 15.73,
+ "tone_ratio": 5.74,
+ "n_outside": 23,
+ "outside": [
+ [
+ "12.7 kHz",
+ 31.3,
+ "above"
+ ],
+ [
+ "10.1 kHz",
+ 30.0,
+ "above"
+ ],
+ [
+ "8 kHz",
+ 26.6,
+ "above"
+ ]
+ ],
+ "attack_ms": 138.0,
+ "t40_ms": 185.0,
+ "active_s": 0.324,
+ "onset_s": 0.061,
+ "env_oct": 0.318,
+ "curve_db": 16.14,
+ "t40_ratio": 1.028,
+ "attack_ratio": 1.373,
+ "active_ratio": 0.729,
+ "ref_median": {
+ "attack_ms": 100.5,
+ "t40_ms": 180.0,
+ "active_s": 0.445
+ },
+ "phrase_z": 3.52,
+ "phrase_min": 3.4,
+ "margin": 2.24,
+ "wins": 1.0,
+ "best_confusion": "a zipper",
+ "as_top": [
+ [
+ "Bang",
+ 0.05
+ ],
+ [
+ "Sound effect",
+ 0.033
+ ],
+ [
+ "Burst, pop",
+ 0.024
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.021,
+ "floor_db": -37.4,
+ "stereo_corr": 0.698,
+ "darken": 0.81,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -72.8,
+ "first_sample": 0.0,
+ "tail_db": -50.7,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity_bench": {
+ "n": 4,
+ "cos_mean": 0.857,
+ "cos_max": 0.918,
+ "tone_min_db": 2.76,
+ "tone_mean_db": 4.07,
+ "near_duplicates": []
+ },
+ "diversity_forge": {
+ "n": 4,
+ "cos_mean": 0.845,
+ "cos_max": 0.931,
+ "tone_min_db": 2.41,
+ "tone_mean_db": 3.38,
+ "near_duplicates": []
+ },
+ "summary": {
+ "made": {},
+ "real_cross": {
+ "n": 4,
+ "typ_z": -10.07,
+ "tone_db": 16.83,
+ "env_oct": 0.43,
+ "curve_db": 14.12,
+ "phrase_z": 2.7,
+ "margin": 0.59,
+ "as_p": 0.11,
+ "tone_ratio": 4.62,
+ "t40_ratio": 0.79,
+ "wins": 1.0,
+ "floor_db": -56.85,
+ "stereo_corr": 0.42,
+ "darken": 0.07,
+ "as_names": 1.0,
+ "artifacts": 0,
+ "var_cos": 0.845
+ },
+ "real_loo": {
+ "n": 4,
+ "typ_z": 1.25,
+ "tone_db": 3.64,
+ "env_oct": 0.22,
+ "curve_db": 10.07,
+ "phrase_z": 3.42,
+ "margin": 2.08,
+ "as_p": 0.03,
+ "tone_ratio": 1.04,
+ "t40_ratio": 1.02,
+ "wins": 1.0,
+ "floor_db": -38.05,
+ "stereo_corr": 0.68,
+ "darken": 0.61,
+ "as_names": 0.25,
+ "artifacts": 0,
+ "var_cos": 0.857
+ }
+ }
+ },
+ "swish": {
+ "n_bench": 3,
+ "n_forge": 7,
+ "note": "",
+ "bench_refs": [
+ "E:/earshot/out/refs/swish/bench/fs-463763_1.wav",
+ "E:/earshot/out/refs/swish/bench/fs-216675_1.wav",
+ "E:/earshot/out/refs/swish/bench/fs-352719_1.wav"
+ ],
+ "made": [],
+ "diversity": {
+ "n": 0
+ },
+ "real_cross": [
+ {
+ "file": "E:/earshot/out/refs/swish/forge/fs-427979_1.wav",
+ "length_s": 0.237,
+ "typ_z": 2.07,
+ "nn_cos": 0.881,
+ "tone_db": 11.08,
+ "tone_ratio": 1.23,
+ "n_outside": 10,
+ "outside": [
+ [
+ "99 Hz",
+ 27.7,
+ "above"
+ ],
+ [
+ "79 Hz",
+ 22.7,
+ "above"
+ ],
+ [
+ "125 Hz",
+ 19.6,
+ "above"
+ ]
+ ],
+ "attack_ms": 25.0,
+ "t40_ms": 110.0,
+ "active_s": 0.134,
+ "onset_s": 0.061,
+ "env_oct": 0.997,
+ "curve_db": 19.87,
+ "t40_ratio": 0.5,
+ "attack_ratio": 0.61,
+ "active_ratio": 0.412,
+ "ref_median": {
+ "attack_ms": 41.0,
+ "t40_ms": 220.0,
+ "active_s": 0.325
+ },
+ "phrase_z": 3.34,
+ "phrase_min": 3.08,
+ "margin": 2.2,
+ "wins": 1.0,
+ "best_confusion": "a sword being drawn",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.174
+ ],
+ [
+ "Grunt",
+ 0.169
+ ],
+ [
+ "Music",
+ 0.041
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.013,
+ "floor_db": -55.6,
+ "stereo_corr": 0.97,
+ "darken": 1.05,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -105.4,
+ "first_sample": 0.0,
+ "tail_db": -69.1,
+ "lead_s": 0.02,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/swish/forge/fs-427979_2.wav",
+ "length_s": 0.417,
+ "typ_z": 1.63,
+ "nn_cos": 0.864,
+ "tone_db": 9.41,
+ "tone_ratio": 1.04,
+ "n_outside": 9,
+ "outside": [
+ [
+ "31 Hz",
+ -23.3,
+ "below"
+ ],
+ [
+ "39 Hz",
+ -20.1,
+ "below"
+ ],
+ [
+ "5.04 kHz",
+ 13.5,
+ "above"
+ ]
+ ],
+ "attack_ms": 78.0,
+ "t40_ms": 145.0,
+ "active_s": 0.247,
+ "onset_s": 0.093,
+ "env_oct": 0.642,
+ "curve_db": 11.38,
+ "t40_ratio": 0.659,
+ "attack_ratio": 1.902,
+ "active_ratio": 0.76,
+ "ref_median": {
+ "attack_ms": 41.0,
+ "t40_ms": 220.0,
+ "active_s": 0.325
+ },
+ "phrase_z": 3.26,
+ "phrase_min": 3.05,
+ "margin": 1.29,
+ "wins": 1.0,
+ "best_confusion": "a sword being drawn",
+ "as_top": [
+ [
+ "Sneeze",
+ 0.095
+ ],
+ [
+ "Zipper (clothing)",
+ 0.094
+ ],
+ [
+ "Whoosh, swoosh, swish",
+ 0.088
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.088,
+ "floor_db": -60.4,
+ "stereo_corr": 0.899,
+ "darken": -0.41,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -85.0,
+ "first_sample": 0.0,
+ "tail_db": -69.1,
+ "lead_s": 0.056,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/swish/forge/fs-427979_3.wav",
+ "length_s": 0.369,
+ "typ_z": 2.4,
+ "nn_cos": 0.899,
+ "tone_db": 13.55,
+ "tone_ratio": 1.5,
+ "n_outside": 13,
+ "outside": [
+ [
+ "50 Hz",
+ -34.5,
+ "below"
+ ],
+ [
+ "39 Hz",
+ -31.5,
+ "below"
+ ],
+ [
+ "31 Hz",
+ -23.3,
+ "below"
+ ]
+ ],
+ "attack_ms": 34.0,
+ "t40_ms": 160.0,
+ "active_s": 0.26,
+ "onset_s": 0.075,
+ "env_oct": 0.35,
+ "curve_db": 12.88,
+ "t40_ratio": 0.727,
+ "attack_ratio": 0.829,
+ "active_ratio": 0.8,
+ "ref_median": {
+ "attack_ms": 41.0,
+ "t40_ms": 220.0,
+ "active_s": 0.325
+ },
+ "phrase_z": 3.16,
+ "phrase_min": 2.87,
+ "margin": 1.15,
+ "wins": 1.0,
+ "best_confusion": "a sword being drawn",
+ "as_top": [
+ [
+ "Sneeze",
+ 0.263
+ ],
+ [
+ "Sound effect",
+ 0.05
+ ],
+ [
+ "Zipper (clothing)",
+ 0.05
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.019,
+ "floor_db": -53.3,
+ "stereo_corr": 0.873,
+ "darken": -0.49,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -103.8,
+ "first_sample": 0.0,
+ "tail_db": -55.9,
+ "lead_s": 0.029,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/swish/forge/fs-427979_4.wav",
+ "length_s": 0.537,
+ "typ_z": 1.94,
+ "nn_cos": 0.882,
+ "tone_db": 10.82,
+ "tone_ratio": 1.2,
+ "n_outside": 10,
+ "outside": [
+ [
+ "39 Hz",
+ -23.4,
+ "below"
+ ],
+ [
+ "31 Hz",
+ -23.1,
+ "below"
+ ],
+ [
+ "10.1 kHz",
+ 22.4,
+ "above"
+ ]
+ ],
+ "attack_ms": 34.0,
+ "t40_ms": 140.0,
+ "active_s": 0.198,
+ "onset_s": 0.072,
+ "env_oct": 0.546,
+ "curve_db": 12.53,
+ "t40_ratio": 0.636,
+ "attack_ratio": 0.829,
+ "active_ratio": 0.609,
+ "ref_median": {
+ "attack_ms": 41.0,
+ "t40_ms": 220.0,
+ "active_s": 0.325
+ },
+ "phrase_z": 3.14,
+ "phrase_min": 2.92,
+ "margin": 1.4,
+ "wins": 1.0,
+ "best_confusion": "a sword being drawn",
+ "as_top": [
+ [
+ "Zipper (clothing)",
+ 0.141
+ ],
+ [
+ "Sound effect",
+ 0.066
+ ],
+ [
+ "Whoosh, swoosh, swish",
+ 0.057
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.057,
+ "floor_db": -64.5,
+ "stereo_corr": 0.866,
+ "darken": -0.38,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -96.3,
+ "first_sample": 0.0,
+ "tail_db": -84.1,
+ "lead_s": 0.031,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/swish/forge/fs-719637_1.wav",
+ "length_s": 0.175,
+ "typ_z": 2.37,
+ "nn_cos": 0.925,
+ "tone_db": 8.56,
+ "tone_ratio": 0.95,
+ "n_outside": 11,
+ "outside": [
+ [
+ "198 Hz",
+ 11.9,
+ "above"
+ ],
+ [
+ "3.17 kHz",
+ 10.6,
+ "above"
+ ],
+ [
+ "2.52 kHz",
+ 10.0,
+ "above"
+ ]
+ ],
+ "attack_ms": 31.0,
+ "t40_ms": 105.0,
+ "active_s": 0.139,
+ "onset_s": 0.011,
+ "env_oct": 0.899,
+ "curve_db": 21.51,
+ "t40_ratio": 0.477,
+ "attack_ratio": 0.756,
+ "active_ratio": 0.428,
+ "ref_median": {
+ "attack_ms": 41.0,
+ "t40_ms": 220.0,
+ "active_s": 0.325
+ },
+ "phrase_z": 3.05,
+ "phrase_min": 2.89,
+ "margin": 1.6,
+ "wins": 1.0,
+ "best_confusion": "a sword being drawn",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.057
+ ],
+ [
+ "Whack, thwack",
+ 0.038
+ ],
+ [
+ "Bang",
+ 0.035
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.003,
+ "floor_db": -49.4,
+ "stereo_corr": 0.816,
+ "darken": -0.16,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -43.3,
+ "first_sample": 0.0,
+ "tail_db": -62.3,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/swish/forge/fs-425856_1.wav",
+ "length_s": 0.453,
+ "typ_z": 0.76,
+ "nn_cos": 0.829,
+ "tone_db": 12.91,
+ "tone_ratio": 1.43,
+ "n_outside": 16,
+ "outside": [
+ [
+ "99 Hz",
+ 22.0,
+ "above"
+ ],
+ [
+ "79 Hz",
+ 21.7,
+ "above"
+ ],
+ [
+ "50 Hz",
+ 21.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 84.0,
+ "t40_ms": 285.0,
+ "active_s": 0.366,
+ "onset_s": 0.079,
+ "env_oct": 0.527,
+ "curve_db": 8.84,
+ "t40_ratio": 1.295,
+ "attack_ratio": 2.049,
+ "active_ratio": 1.126,
+ "ref_median": {
+ "attack_ms": 41.0,
+ "t40_ms": 220.0,
+ "active_s": 0.325
+ },
+ "phrase_z": 3.68,
+ "phrase_min": 3.54,
+ "margin": 2.79,
+ "wins": 1.0,
+ "best_confusion": "a sword being drawn",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.152
+ ],
+ [
+ "Whoosh, swoosh, swish",
+ 0.151
+ ],
+ [
+ "Slam",
+ 0.057
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.151,
+ "floor_db": -43.7,
+ "stereo_corr": -0.161,
+ "darken": 0.26,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -73.2,
+ "first_sample": 0.0,
+ "tail_db": -40.3,
+ "lead_s": 0.029,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/swish/forge/fs-425856_2.wav",
+ "length_s": 0.311,
+ "typ_z": -0.79,
+ "nn_cos": 0.761,
+ "tone_db": 15.01,
+ "tone_ratio": 1.67,
+ "n_outside": 25,
+ "outside": [
+ [
+ "31 Hz",
+ 33.0,
+ "above"
+ ],
+ [
+ "39 Hz",
+ 24.1,
+ "above"
+ ],
+ [
+ "12.7 kHz",
+ -20.9,
+ "below"
+ ]
+ ],
+ "attack_ms": 87.0,
+ "t40_ms": 110.0,
+ "active_s": 0.195,
+ "onset_s": 0.1,
+ "env_oct": 0.941,
+ "curve_db": 17.6,
+ "t40_ratio": 0.5,
+ "attack_ratio": 2.122,
+ "active_ratio": 0.6,
+ "ref_median": {
+ "attack_ms": 41.0,
+ "t40_ms": 220.0,
+ "active_s": 0.325
+ },
+ "phrase_z": 3.22,
+ "phrase_min": 3.16,
+ "margin": 2.18,
+ "wins": 1.0,
+ "best_confusion": "a sword being drawn",
+ "as_top": [
+ [
+ "Bang",
+ 0.104
+ ],
+ [
+ "Sound effect",
+ 0.077
+ ],
+ [
+ "Whoosh, swoosh, swish",
+ 0.041
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.041,
+ "floor_db": -63.7,
+ "stereo_corr": 0.225,
+ "darken": 0.08,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -81.0,
+ "first_sample": 0.0,
+ "tail_db": -47.2,
+ "lead_s": 0.057,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_loo": [
+ {
+ "file": "E:/earshot/out/refs/swish/bench/fs-463763_1.wav",
+ "length_s": 0.398,
+ "nn_cos": 0.745,
+ "tone_db": 8.6,
+ "tone_ratio": 0.92,
+ "n_outside": 18,
+ "outside": [
+ [
+ "1 kHz",
+ 21.6,
+ "above"
+ ],
+ [
+ "397 Hz",
+ -12.9,
+ "below"
+ ],
+ [
+ "794 Hz",
+ 11.6,
+ "above"
+ ]
+ ],
+ "attack_ms": 108.0,
+ "t40_ms": 270.0,
+ "active_s": 0.347,
+ "onset_s": 0.033,
+ "env_oct": 0.908,
+ "curve_db": 16.45,
+ "t40_ratio": 1.636,
+ "attack_ratio": 2.919,
+ "active_ratio": 1.382,
+ "ref_median": {
+ "attack_ms": 37.0,
+ "t40_ms": 165.0,
+ "active_s": 0.251
+ },
+ "phrase_z": 3.61,
+ "phrase_min": 3.31,
+ "margin": 2.36,
+ "wins": 1.0,
+ "best_confusion": "a sword being drawn",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.138
+ ],
+ [
+ "Sound effect",
+ 0.063
+ ],
+ [
+ "Music",
+ 0.033
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.138,
+ "floor_db": -28.7,
+ "stereo_corr": 0.991,
+ "darken": -0.18,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -51.9,
+ "first_sample": 0.0,
+ "tail_db": -49.3,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/swish/bench/fs-216675_1.wav",
+ "length_s": 0.23,
+ "nn_cos": 0.908,
+ "tone_db": 9.63,
+ "tone_ratio": 1.09,
+ "n_outside": 12,
+ "outside": [
+ [
+ "50 Hz",
+ -28.6,
+ "below"
+ ],
+ [
+ "39 Hz",
+ -23.6,
+ "below"
+ ],
+ [
+ "250 Hz",
+ 11.3,
+ "above"
+ ]
+ ],
+ "attack_ms": 33.0,
+ "t40_ms": 110.0,
+ "active_s": 0.177,
+ "onset_s": 0.023,
+ "env_oct": 1.085,
+ "curve_db": 21.51,
+ "t40_ratio": 0.449,
+ "attack_ratio": 0.443,
+ "active_ratio": 0.527,
+ "ref_median": {
+ "attack_ms": 74.5,
+ "t40_ms": 245.0,
+ "active_s": 0.336
+ },
+ "phrase_z": 3.36,
+ "phrase_min": 3.1,
+ "margin": 1.76,
+ "wins": 1.0,
+ "best_confusion": "a sword being drawn",
+ "as_top": [
+ [
+ "Music",
+ 0.209
+ ],
+ [
+ "Sound effect",
+ 0.206
+ ],
+ [
+ "Bang",
+ 0.138
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.005,
+ "floor_db": -46.7,
+ "stereo_corr": 1.0,
+ "darken": -0.77,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -98.6,
+ "first_sample": 0.0,
+ "tail_db": -63.8,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/swish/bench/fs-352719_1.wav",
+ "length_s": 0.435,
+ "nn_cos": 0.908,
+ "tone_db": 9.01,
+ "tone_ratio": 0.99,
+ "n_outside": 7,
+ "outside": [
+ [
+ "31 Hz",
+ 25.8,
+ "above"
+ ],
+ [
+ "39 Hz",
+ 23.0,
+ "above"
+ ],
+ [
+ "50 Hz",
+ 19.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 41.0,
+ "t40_ms": 220.0,
+ "active_s": 0.325,
+ "onset_s": 0.065,
+ "env_oct": 0.435,
+ "curve_db": 8.63,
+ "t40_ratio": 1.158,
+ "attack_ratio": 0.582,
+ "active_ratio": 1.24,
+ "ref_median": {
+ "attack_ms": 70.5,
+ "t40_ms": 190.0,
+ "active_s": 0.262
+ },
+ "phrase_z": 3.5,
+ "phrase_min": 3.01,
+ "margin": 1.92,
+ "wins": 1.0,
+ "best_confusion": "a sword being drawn",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.102
+ ],
+ [
+ "Sound effect",
+ 0.08
+ ],
+ [
+ "Zipper (clothing)",
+ 0.035
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.102,
+ "floor_db": -44.6,
+ "stereo_corr": 1.0,
+ "darken": -0.45,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -54.7,
+ "first_sample": 0.0,
+ "tail_db": -64.7,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_rev": [
+ {
+ "file": "E:/earshot/out/refs/swish/bench/fs-463763_1.wav",
+ "length_s": 0.398,
+ "typ_z": -2.23,
+ "nn_cos": 0.791,
+ "tone_db": 9.88,
+ "tone_ratio": 0.8,
+ "n_outside": 5,
+ "outside": [
+ [
+ "1 kHz",
+ 20.5,
+ "above"
+ ],
+ [
+ "794 Hz",
+ 13.8,
+ "above"
+ ],
+ [
+ "630 Hz",
+ 13.2,
+ "above"
+ ]
+ ],
+ "attack_ms": 108.0,
+ "t40_ms": 270.0,
+ "active_s": 0.347,
+ "onset_s": 0.033,
+ "env_oct": 1.141,
+ "curve_db": 17.34,
+ "t40_ratio": 1.929,
+ "attack_ratio": 3.176,
+ "active_ratio": 1.753,
+ "ref_median": {
+ "attack_ms": 34.0,
+ "t40_ms": 140.0,
+ "active_s": 0.198
+ },
+ "phrase_z": 3.61,
+ "phrase_min": 3.31,
+ "margin": 2.36,
+ "wins": 1.0,
+ "best_confusion": "a sword being drawn",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.138
+ ],
+ [
+ "Sound effect",
+ 0.063
+ ],
+ [
+ "Music",
+ 0.033
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.138,
+ "floor_db": -28.7,
+ "stereo_corr": 0.991,
+ "darken": -0.18,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -51.9,
+ "first_sample": 0.0,
+ "tail_db": -49.3,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/swish/bench/fs-216675_1.wav",
+ "length_s": 0.23,
+ "typ_z": 0.54,
+ "nn_cos": 0.925,
+ "tone_db": 7.28,
+ "tone_ratio": 0.59,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 33.0,
+ "t40_ms": 110.0,
+ "active_s": 0.177,
+ "onset_s": 0.023,
+ "env_oct": 0.184,
+ "curve_db": 13.98,
+ "t40_ratio": 0.786,
+ "attack_ratio": 0.971,
+ "active_ratio": 0.894,
+ "ref_median": {
+ "attack_ms": 34.0,
+ "t40_ms": 140.0,
+ "active_s": 0.198
+ },
+ "phrase_z": 3.36,
+ "phrase_min": 3.1,
+ "margin": 1.76,
+ "wins": 1.0,
+ "best_confusion": "a sword being drawn",
+ "as_top": [
+ [
+ "Music",
+ 0.209
+ ],
+ [
+ "Sound effect",
+ 0.206
+ ],
+ [
+ "Bang",
+ 0.138
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.005,
+ "floor_db": -46.7,
+ "stereo_corr": 1.0,
+ "darken": -0.77,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -98.6,
+ "first_sample": 0.0,
+ "tail_db": -63.8,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/swish/bench/fs-352719_1.wav",
+ "length_s": 0.435,
+ "typ_z": 0.3,
+ "nn_cos": 0.899,
+ "tone_db": 9.1,
+ "tone_ratio": 0.74,
+ "n_outside": 3,
+ "outside": [
+ [
+ "250 Hz",
+ -14.0,
+ "below"
+ ],
+ [
+ "500 Hz",
+ 13.9,
+ "above"
+ ],
+ [
+ "198 Hz",
+ -7.7,
+ "below"
+ ]
+ ],
+ "attack_ms": 41.0,
+ "t40_ms": 220.0,
+ "active_s": 0.325,
+ "onset_s": 0.065,
+ "env_oct": 0.546,
+ "curve_db": 13.48,
+ "t40_ratio": 1.571,
+ "attack_ratio": 1.206,
+ "active_ratio": 1.641,
+ "ref_median": {
+ "attack_ms": 34.0,
+ "t40_ms": 140.0,
+ "active_s": 0.198
+ },
+ "phrase_z": 3.5,
+ "phrase_min": 3.01,
+ "margin": 1.92,
+ "wins": 1.0,
+ "best_confusion": "a sword being drawn",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.102
+ ],
+ [
+ "Sound effect",
+ 0.08
+ ],
+ [
+ "Zipper (clothing)",
+ 0.035
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.102,
+ "floor_db": -44.6,
+ "stereo_corr": 1.0,
+ "darken": -0.45,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -54.7,
+ "first_sample": 0.0,
+ "tail_db": -64.7,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity_bench": {
+ "n": 3,
+ "cos_mean": 0.79,
+ "cos_max": 0.908,
+ "tone_min_db": 9.83,
+ "tone_mean_db": 10.48,
+ "near_duplicates": []
+ },
+ "diversity_forge": {
+ "n": 7,
+ "cos_mean": 0.811,
+ "cos_max": 0.956,
+ "tone_min_db": 5.43,
+ "tone_mean_db": 15.54,
+ "near_duplicates": []
+ },
+ "summary": {
+ "made": {},
+ "real_cross": {
+ "n": 7,
+ "typ_z": 1.94,
+ "tone_db": 11.08,
+ "env_oct": 0.64,
+ "curve_db": 12.88,
+ "phrase_z": 3.22,
+ "margin": 1.6,
+ "as_p": 0.04,
+ "tone_ratio": 1.23,
+ "t40_ratio": 0.64,
+ "wins": 1.0,
+ "floor_db": -55.6,
+ "stereo_corr": 0.87,
+ "darken": -0.16,
+ "as_names": 0.57,
+ "artifacts": 0,
+ "var_cos": 0.811
+ },
+ "real_loo": {
+ "n": 3,
+ "tone_db": 9.01,
+ "env_oct": 0.91,
+ "curve_db": 16.45,
+ "phrase_z": 3.5,
+ "margin": 1.92,
+ "as_p": 0.1,
+ "tone_ratio": 0.99,
+ "t40_ratio": 1.16,
+ "wins": 1.0,
+ "floor_db": -44.6,
+ "stereo_corr": 1.0,
+ "darken": -0.45,
+ "as_names": 0.67,
+ "artifacts": 0,
+ "var_cos": 0.79
+ }
+ }
+ },
+ "tap": {
+ "n_bench": 6,
+ "n_forge": 7,
+ "note": "",
+ "bench_refs": [
+ "E:/earshot/out/refs/tap/bench/fs-631822_1.wav",
+ "E:/earshot/out/refs/tap/bench/fs-631822_2.wav",
+ "E:/earshot/out/refs/tap/bench/fs-631822_3.wav",
+ "E:/earshot/out/refs/tap/bench/fs-707827_1.wav",
+ "E:/earshot/out/refs/tap/bench/fs-707827_2.wav",
+ "E:/earshot/out/refs/tap/bench/fs-707827_3.wav"
+ ],
+ "made": [],
+ "diversity": {
+ "n": 0
+ },
+ "real_cross": [
+ {
+ "file": "E:/earshot/out/refs/tap/forge/kit-ui_click_01.wav",
+ "length_s": 0.096,
+ "typ_z": -0.53,
+ "nn_cos": 0.825,
+ "tone_db": 8.05,
+ "tone_ratio": 0.48,
+ "n_outside": 6,
+ "outside": [
+ [
+ "39 Hz",
+ 23.6,
+ "above"
+ ],
+ [
+ "31 Hz",
+ 20.0,
+ "above"
+ ],
+ [
+ "4 kHz",
+ -14.8,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 60.0,
+ "active_s": 0.082,
+ "onset_s": 0.008,
+ "env_oct": 1.587,
+ "curve_db": 12.17,
+ "t40_ratio": 1.091,
+ "attack_ratio": 0.05,
+ "active_ratio": 0.804,
+ "ref_median": {
+ "attack_ms": 20.0,
+ "t40_ms": 55.0,
+ "active_s": 0.102
+ },
+ "phrase_z": 2.21,
+ "phrase_min": 1.8,
+ "margin": 0.77,
+ "wins": 0.5,
+ "best_confusion": "a keyboard key",
+ "as_top": [
+ [
+ "Speech",
+ 0.125
+ ],
+ [
+ "Vehicle",
+ 0.101
+ ],
+ [
+ "Music",
+ 0.094
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -43.6,
+ "stereo_corr": 1.0,
+ "darken": 1.51,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -52.9,
+ "first_sample": 0.0,
+ "tail_db": -39.4,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/tap/forge/fs-704862_1.wav",
+ "length_s": 0.131,
+ "typ_z": -1.17,
+ "nn_cos": 0.743,
+ "tone_db": 10.99,
+ "tone_ratio": 0.65,
+ "n_outside": 7,
+ "outside": [
+ [
+ "39 Hz",
+ 25.8,
+ "above"
+ ],
+ [
+ "31 Hz",
+ 21.9,
+ "above"
+ ],
+ [
+ "2.52 kHz",
+ -21.2,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 75.0,
+ "active_s": 0.101,
+ "onset_s": 0.009,
+ "env_oct": 1.595,
+ "curve_db": 9.98,
+ "t40_ratio": 1.364,
+ "attack_ratio": 0.05,
+ "active_ratio": 0.99,
+ "ref_median": {
+ "attack_ms": 20.0,
+ "t40_ms": 55.0,
+ "active_s": 0.102
+ },
+ "phrase_z": 1.99,
+ "phrase_min": 1.77,
+ "margin": 0.93,
+ "wins": 0.0,
+ "best_confusion": "a knock on a door",
+ "as_top": [
+ [
+ "Music",
+ 0.473
+ ],
+ [
+ "Speech",
+ 0.082
+ ],
+ [
+ "Sound effect",
+ 0.053
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -43.8,
+ "stereo_corr": 1.0,
+ "darken": 2.13,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -64.9,
+ "first_sample": 0.0,
+ "tail_db": -55.0,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/tap/forge/fs-704862_2.wav",
+ "length_s": 0.131,
+ "typ_z": -1.04,
+ "nn_cos": 0.762,
+ "tone_db": 9.55,
+ "tone_ratio": 0.57,
+ "n_outside": 6,
+ "outside": [
+ [
+ "39 Hz",
+ 22.1,
+ "above"
+ ],
+ [
+ "31 Hz",
+ 19.2,
+ "above"
+ ],
+ [
+ "3.17 kHz",
+ -14.6,
+ "below"
+ ]
+ ],
+ "attack_ms": 2.0,
+ "t40_ms": 90.0,
+ "active_s": 0.091,
+ "onset_s": 0.029,
+ "env_oct": 1.399,
+ "curve_db": 13.67,
+ "t40_ratio": 1.636,
+ "attack_ratio": 0.1,
+ "active_ratio": 0.892,
+ "ref_median": {
+ "attack_ms": 20.0,
+ "t40_ms": 55.0,
+ "active_s": 0.102
+ },
+ "phrase_z": 1.92,
+ "phrase_min": 1.55,
+ "margin": 0.93,
+ "wins": 0.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Music",
+ 0.204
+ ],
+ [
+ "Speech",
+ 0.113
+ ],
+ [
+ "Vehicle",
+ 0.04
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.005,
+ "floor_db": -35.7,
+ "stereo_corr": 1.0,
+ "darken": 2.2,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -60.2,
+ "first_sample": 0.0,
+ "tail_db": -47.4,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/tap/forge/fs-704862_3.wav",
+ "length_s": 0.148,
+ "typ_z": -0.79,
+ "nn_cos": 0.774,
+ "tone_db": 8.15,
+ "tone_ratio": 0.49,
+ "n_outside": 5,
+ "outside": [
+ [
+ "39 Hz",
+ 23.2,
+ "above"
+ ],
+ [
+ "31 Hz",
+ 19.9,
+ "above"
+ ],
+ [
+ "3.17 kHz",
+ -13.6,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 50.0,
+ "active_s": 0.05,
+ "onset_s": 0.08,
+ "env_oct": 1.829,
+ "curve_db": 12.85,
+ "t40_ratio": 0.909,
+ "attack_ratio": 0.05,
+ "active_ratio": 0.49,
+ "ref_median": {
+ "attack_ms": 20.0,
+ "t40_ms": 55.0,
+ "active_s": 0.102
+ },
+ "phrase_z": 1.85,
+ "phrase_min": 1.26,
+ "margin": 0.93,
+ "wins": 0.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Chopping (food)",
+ 0.086
+ ],
+ [
+ "Sound effect",
+ 0.028
+ ],
+ [
+ "Speech",
+ 0.024
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.019,
+ "floor_db": -47.7,
+ "stereo_corr": 1.0,
+ "darken": 1.31,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -69.0,
+ "first_sample": 0.0,
+ "tail_db": -53.3,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/tap/forge/fs-452621_1.wav",
+ "length_s": 0.139,
+ "typ_z": -1.67,
+ "nn_cos": 0.721,
+ "tone_db": 9.21,
+ "tone_ratio": 0.55,
+ "n_outside": 2,
+ "outside": [
+ [
+ "39 Hz",
+ 33.5,
+ "above"
+ ],
+ [
+ "31 Hz",
+ 16.5,
+ "above"
+ ]
+ ],
+ "attack_ms": 6.0,
+ "t40_ms": 110.0,
+ "active_s": 0.125,
+ "onset_s": 0.005,
+ "env_oct": 1.01,
+ "curve_db": 20.83,
+ "t40_ratio": 2.0,
+ "attack_ratio": 0.3,
+ "active_ratio": 1.225,
+ "ref_median": {
+ "attack_ms": 20.0,
+ "t40_ms": 55.0,
+ "active_s": 0.102
+ },
+ "phrase_z": 1.35,
+ "phrase_min": 1.05,
+ "margin": 0.61,
+ "wins": 0.0,
+ "best_confusion": "a knock on a door",
+ "as_top": [
+ [
+ "Bang",
+ 0.232
+ ],
+ [
+ "Sound effect",
+ 0.128
+ ],
+ [
+ "Slam",
+ 0.089
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.012,
+ "floor_db": -38.1,
+ "stereo_corr": 1.0,
+ "darken": -0.67,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -50.1,
+ "first_sample": 0.0,
+ "tail_db": -42.5,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/tap/forge/fs-452621_2.wav",
+ "length_s": 0.216,
+ "typ_z": -2.67,
+ "nn_cos": 0.616,
+ "tone_db": 9.35,
+ "tone_ratio": 0.56,
+ "n_outside": 2,
+ "outside": [
+ [
+ "39 Hz",
+ 25.5,
+ "above"
+ ],
+ [
+ "50 Hz",
+ 10.2,
+ "above"
+ ]
+ ],
+ "attack_ms": 2.0,
+ "t40_ms": 110.0,
+ "active_s": 0.11,
+ "onset_s": 0.085,
+ "env_oct": 1.477,
+ "curve_db": 21.54,
+ "t40_ratio": 2.0,
+ "attack_ratio": 0.1,
+ "active_ratio": 1.078,
+ "ref_median": {
+ "attack_ms": 20.0,
+ "t40_ms": 55.0,
+ "active_s": 0.102
+ },
+ "phrase_z": 1.63,
+ "phrase_min": 1.23,
+ "margin": 0.25,
+ "wins": 0.0,
+ "best_confusion": "a knock on a door",
+ "as_top": [
+ [
+ "Knock",
+ 0.382
+ ],
+ [
+ "Sound effect",
+ 0.068
+ ],
+ [
+ "Speech",
+ 0.053
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.382,
+ "floor_db": -72.0,
+ "stereo_corr": 1.0,
+ "darken": -0.64,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -40.6,
+ "first_sample": 0.0,
+ "tail_db": -53.4,
+ "lead_s": 0.068,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/tap/forge/fs-452621_3.wav",
+ "length_s": 0.111,
+ "typ_z": -1.36,
+ "nn_cos": 0.716,
+ "tone_db": 9.99,
+ "tone_ratio": 0.6,
+ "n_outside": 2,
+ "outside": [
+ [
+ "39 Hz",
+ 23.6,
+ "above"
+ ],
+ [
+ "50 Hz",
+ 12.7,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 100.0,
+ "active_s": 0.1,
+ "onset_s": 0.005,
+ "env_oct": 1.738,
+ "curve_db": 20.15,
+ "t40_ratio": 1.818,
+ "attack_ratio": 0.05,
+ "active_ratio": 0.98,
+ "ref_median": {
+ "attack_ms": 20.0,
+ "t40_ms": 55.0,
+ "active_s": 0.102
+ },
+ "phrase_z": 1.61,
+ "phrase_min": 1.18,
+ "margin": 0.92,
+ "wins": 0.0,
+ "best_confusion": "a keyboard key",
+ "as_top": [
+ [
+ "Music",
+ 0.316
+ ],
+ [
+ "Sound effect",
+ 0.093
+ ],
+ [
+ "Speech",
+ 0.091
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.009,
+ "floor_db": -39.6,
+ "stereo_corr": 1.0,
+ "darken": -0.68,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -39.1,
+ "first_sample": 0.0,
+ "tail_db": -41.6,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_loo": [
+ {
+ "file": "E:/earshot/out/refs/tap/bench/fs-631822_1.wav",
+ "length_s": 0.143,
+ "typ_z": -1.92,
+ "nn_cos": 0.757,
+ "tone_db": 18.84,
+ "tone_ratio": 1.14,
+ "n_outside": 5,
+ "outside": [
+ [
+ "794 Hz",
+ 25.3,
+ "above"
+ ],
+ [
+ "1 kHz",
+ 20.2,
+ "above"
+ ],
+ [
+ "2.52 kHz",
+ 15.9,
+ "above"
+ ]
+ ],
+ "attack_ms": 31.0,
+ "t40_ms": 105.0,
+ "active_s": 0.135,
+ "onset_s": 0.005,
+ "env_oct": 1.228,
+ "curve_db": 24.18,
+ "t40_ratio": 1.909,
+ "attack_ratio": 3.444,
+ "active_ratio": 1.957,
+ "ref_median": {
+ "attack_ms": 9.0,
+ "t40_ms": 55.0,
+ "active_s": 0.069
+ },
+ "phrase_z": 0.43,
+ "phrase_min": -0.2,
+ "margin": 0.08,
+ "wins": 0.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Rain on surface",
+ 0.09
+ ],
+ [
+ "Speech",
+ 0.08
+ ],
+ [
+ "Rain",
+ 0.073
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.004,
+ "floor_db": -23.7,
+ "stereo_corr": 1.0,
+ "darken": 0.13,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -73.1,
+ "first_sample": 0.0,
+ "tail_db": -33.0,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/tap/bench/fs-631822_2.wav",
+ "length_s": 0.151,
+ "typ_z": 0.47,
+ "nn_cos": 0.879,
+ "tone_db": 14.41,
+ "tone_ratio": 0.85,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 98.0,
+ "t40_ms": 40.0,
+ "active_s": 0.135,
+ "onset_s": 0.005,
+ "env_oct": 1.624,
+ "curve_db": 26.68,
+ "t40_ratio": 0.727,
+ "attack_ratio": 10.889,
+ "active_ratio": 1.957,
+ "ref_median": {
+ "attack_ms": 9.0,
+ "t40_ms": 55.0,
+ "active_s": 0.069
+ },
+ "phrase_z": 1.74,
+ "phrase_min": 1.24,
+ "margin": 0.88,
+ "wins": 0.0,
+ "best_confusion": "a keyboard key",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.063
+ ],
+ [
+ "Knock",
+ 0.038
+ ],
+ [
+ "Music",
+ 0.037
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.038,
+ "floor_db": -35.9,
+ "stereo_corr": 1.0,
+ "darken": 0.93,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -63.9,
+ "first_sample": 0.0,
+ "tail_db": -47.6,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/tap/bench/fs-631822_3.wav",
+ "length_s": 0.2,
+ "typ_z": -0.52,
+ "nn_cos": 0.879,
+ "tone_db": 18.41,
+ "tone_ratio": 1.11,
+ "n_outside": 6,
+ "outside": [
+ [
+ "500 Hz",
+ 25.8,
+ "above"
+ ],
+ [
+ "397 Hz",
+ 22.6,
+ "above"
+ ],
+ [
+ "5.04 kHz",
+ 15.3,
+ "above"
+ ]
+ ],
+ "attack_ms": 115.0,
+ "t40_ms": 165.0,
+ "active_s": 0.175,
+ "onset_s": 0.015,
+ "env_oct": 2.201,
+ "curve_db": 21.11,
+ "t40_ratio": 3.0,
+ "attack_ratio": 12.778,
+ "active_ratio": 2.536,
+ "ref_median": {
+ "attack_ms": 9.0,
+ "t40_ms": 55.0,
+ "active_s": 0.069
+ },
+ "phrase_z": 1.75,
+ "phrase_min": 1.12,
+ "margin": 0.31,
+ "wins": 0.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Speech",
+ 0.067
+ ],
+ [
+ "Animal",
+ 0.047
+ ],
+ [
+ "Music",
+ 0.039
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.012,
+ "floor_db": -36.4,
+ "stereo_corr": 1.0,
+ "darken": 1.48,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -96.5,
+ "first_sample": 0.0,
+ "tail_db": -46.5,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/tap/bench/fs-707827_1.wav",
+ "length_s": 0.087,
+ "typ_z": 3.68,
+ "nn_cos": 0.98,
+ "tone_db": 17.02,
+ "tone_ratio": 1.03,
+ "n_outside": 2,
+ "outside": [
+ [
+ "31 Hz",
+ 12.1,
+ "above"
+ ],
+ [
+ "50 Hz",
+ 5.9,
+ "above"
+ ]
+ ],
+ "attack_ms": 9.0,
+ "t40_ms": 55.0,
+ "active_s": 0.064,
+ "onset_s": 0.006,
+ "env_oct": 0.954,
+ "curve_db": 2.07,
+ "t40_ratio": 1.0,
+ "attack_ratio": 0.29,
+ "active_ratio": 0.474,
+ "ref_median": {
+ "attack_ms": 31.0,
+ "t40_ms": 55.0,
+ "active_s": 0.135
+ },
+ "phrase_z": 1.46,
+ "phrase_min": 0.78,
+ "margin": 0.4,
+ "wins": 0.0,
+ "best_confusion": "a knock on a door",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.084
+ ],
+ [
+ "Music",
+ 0.072
+ ],
+ [
+ "Speech",
+ 0.056
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.007,
+ "floor_db": -43.5,
+ "stereo_corr": 1.0,
+ "darken": 0.47,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -65.8,
+ "first_sample": 0.0,
+ "tail_db": -53.2,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/tap/bench/fs-707827_2.wav",
+ "length_s": 0.083,
+ "typ_z": 2.85,
+ "nn_cos": 0.949,
+ "tone_db": 16.32,
+ "tone_ratio": 0.96,
+ "n_outside": 3,
+ "outside": [
+ [
+ "630 Hz",
+ -29.1,
+ "below"
+ ],
+ [
+ "31 Hz",
+ -25.1,
+ "below"
+ ],
+ [
+ "500 Hz",
+ -22.9,
+ "below"
+ ]
+ ],
+ "attack_ms": 6.0,
+ "t40_ms": 50.0,
+ "active_s": 0.069,
+ "onset_s": 0.006,
+ "env_oct": 1.158,
+ "curve_db": 1.91,
+ "t40_ratio": 0.909,
+ "attack_ratio": 0.194,
+ "active_ratio": 0.511,
+ "ref_median": {
+ "attack_ms": 31.0,
+ "t40_ms": 55.0,
+ "active_s": 0.135
+ },
+ "phrase_z": 1.7,
+ "phrase_min": 1.03,
+ "margin": 0.18,
+ "wins": 0.0,
+ "best_confusion": "a knock on a door",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.073
+ ],
+ [
+ "Music",
+ 0.055
+ ],
+ [
+ "Speech",
+ 0.047
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.005,
+ "floor_db": -40.4,
+ "stereo_corr": 1.0,
+ "darken": 2.29,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -72.6,
+ "first_sample": 0.0,
+ "tail_db": -50.7,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/tap/bench/fs-707827_3.wav",
+ "length_s": 0.088,
+ "typ_z": 3.57,
+ "nn_cos": 0.98,
+ "tone_db": 16.54,
+ "tone_ratio": 0.97,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 8.0,
+ "t40_ms": 55.0,
+ "active_s": 0.064,
+ "onset_s": 0.006,
+ "env_oct": 1.01,
+ "curve_db": 1.94,
+ "t40_ratio": 1.0,
+ "attack_ratio": 0.258,
+ "active_ratio": 0.474,
+ "ref_median": {
+ "attack_ms": 31.0,
+ "t40_ms": 55.0,
+ "active_s": 0.135
+ },
+ "phrase_z": 1.55,
+ "phrase_min": 0.82,
+ "margin": 0.51,
+ "wins": 0.0,
+ "best_confusion": "a knock on a door",
+ "as_top": [
+ [
+ "Music",
+ 0.091
+ ],
+ [
+ "Speech",
+ 0.077
+ ],
+ [
+ "Sound effect",
+ 0.068
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.005,
+ "floor_db": -43.0,
+ "stereo_corr": 1.0,
+ "darken": 0.94,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -87.6,
+ "first_sample": 0.0,
+ "tail_db": -53.0,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_rev": [
+ {
+ "file": "E:/earshot/out/refs/tap/bench/fs-631822_1.wav",
+ "length_s": 0.143,
+ "typ_z": -3.52,
+ "nn_cos": 0.702,
+ "tone_db": 14.26,
+ "tone_ratio": 2.21,
+ "n_outside": 15,
+ "outside": [
+ [
+ "794 Hz",
+ 23.1,
+ "above"
+ ],
+ [
+ "3.17 kHz",
+ 23.0,
+ "above"
+ ],
+ [
+ "1.26 kHz",
+ 22.0,
+ "above"
+ ]
+ ],
+ "attack_ms": 31.0,
+ "t40_ms": 105.0,
+ "active_s": 0.135,
+ "onset_s": 0.005,
+ "env_oct": 1.87,
+ "curve_db": 15.96,
+ "t40_ratio": 1.167,
+ "attack_ratio": 31.0,
+ "active_ratio": 1.35,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 90.0,
+ "active_s": 0.1
+ },
+ "phrase_z": 0.43,
+ "phrase_min": -0.2,
+ "margin": 0.08,
+ "wins": 0.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Rain on surface",
+ 0.09
+ ],
+ [
+ "Speech",
+ 0.08
+ ],
+ [
+ "Rain",
+ 0.073
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.004,
+ "floor_db": -23.7,
+ "stereo_corr": 1.0,
+ "darken": 0.13,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -73.1,
+ "first_sample": 0.0,
+ "tail_db": -33.0,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/tap/bench/fs-631822_2.wav",
+ "length_s": 0.151,
+ "typ_z": -0.76,
+ "nn_cos": 0.825,
+ "tone_db": 10.92,
+ "tone_ratio": 1.69,
+ "n_outside": 9,
+ "outside": [
+ [
+ "39 Hz",
+ -20.8,
+ "below"
+ ],
+ [
+ "1.26 kHz",
+ 19.7,
+ "above"
+ ],
+ [
+ "50 Hz",
+ -14.6,
+ "below"
+ ]
+ ],
+ "attack_ms": 98.0,
+ "t40_ms": 40.0,
+ "active_s": 0.135,
+ "onset_s": 0.005,
+ "env_oct": 2.739,
+ "curve_db": 28.76,
+ "t40_ratio": 0.444,
+ "attack_ratio": 98.0,
+ "active_ratio": 1.35,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 90.0,
+ "active_s": 0.1
+ },
+ "phrase_z": 1.74,
+ "phrase_min": 1.24,
+ "margin": 0.88,
+ "wins": 0.0,
+ "best_confusion": "a keyboard key",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.063
+ ],
+ [
+ "Knock",
+ 0.038
+ ],
+ [
+ "Music",
+ 0.037
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.038,
+ "floor_db": -35.9,
+ "stereo_corr": 1.0,
+ "darken": 0.93,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -63.9,
+ "first_sample": 0.0,
+ "tail_db": -47.6,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/tap/bench/fs-631822_3.wav",
+ "length_s": 0.2,
+ "typ_z": -1.46,
+ "nn_cos": 0.774,
+ "tone_db": 14.52,
+ "tone_ratio": 2.25,
+ "n_outside": 17,
+ "outside": [
+ [
+ "39 Hz",
+ -26.5,
+ "below"
+ ],
+ [
+ "10.1 kHz",
+ 23.2,
+ "above"
+ ],
+ [
+ "500 Hz",
+ 21.6,
+ "above"
+ ]
+ ],
+ "attack_ms": 115.0,
+ "t40_ms": 165.0,
+ "active_s": 0.175,
+ "onset_s": 0.015,
+ "env_oct": 2.842,
+ "curve_db": 20.16,
+ "t40_ratio": 1.833,
+ "attack_ratio": 115.0,
+ "active_ratio": 1.75,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 90.0,
+ "active_s": 0.1
+ },
+ "phrase_z": 1.75,
+ "phrase_min": 1.12,
+ "margin": 0.31,
+ "wins": 0.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Speech",
+ 0.067
+ ],
+ [
+ "Animal",
+ 0.047
+ ],
+ [
+ "Music",
+ 0.039
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.012,
+ "floor_db": -36.4,
+ "stereo_corr": 1.0,
+ "darken": 1.48,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -96.5,
+ "first_sample": 0.0,
+ "tail_db": -46.5,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/tap/bench/fs-707827_1.wav",
+ "length_s": 0.087,
+ "typ_z": -3.03,
+ "nn_cos": 0.711,
+ "tone_db": 13.6,
+ "tone_ratio": 2.11,
+ "n_outside": 8,
+ "outside": [
+ [
+ "39 Hz",
+ -46.2,
+ "below"
+ ],
+ [
+ "12.7 kHz",
+ -21.1,
+ "below"
+ ],
+ [
+ "16 kHz",
+ -20.1,
+ "below"
+ ]
+ ],
+ "attack_ms": 9.0,
+ "t40_ms": 55.0,
+ "active_s": 0.064,
+ "onset_s": 0.006,
+ "env_oct": 1.508,
+ "curve_db": 10.59,
+ "t40_ratio": 0.611,
+ "attack_ratio": 9.0,
+ "active_ratio": 0.64,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 90.0,
+ "active_s": 0.1
+ },
+ "phrase_z": 1.46,
+ "phrase_min": 0.78,
+ "margin": 0.4,
+ "wins": 0.0,
+ "best_confusion": "a knock on a door",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.084
+ ],
+ [
+ "Music",
+ 0.072
+ ],
+ [
+ "Speech",
+ 0.056
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.007,
+ "floor_db": -43.5,
+ "stereo_corr": 1.0,
+ "darken": 0.47,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -65.8,
+ "first_sample": 0.0,
+ "tail_db": -53.2,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/tap/bench/fs-707827_2.wav",
+ "length_s": 0.083,
+ "typ_z": -3.06,
+ "nn_cos": 0.711,
+ "tone_db": 12.82,
+ "tone_ratio": 1.99,
+ "n_outside": 10,
+ "outside": [
+ [
+ "12.7 kHz",
+ -21.1,
+ "below"
+ ],
+ [
+ "16 kHz",
+ -20.1,
+ "below"
+ ],
+ [
+ "8 kHz",
+ -18.9,
+ "below"
+ ]
+ ],
+ "attack_ms": 6.0,
+ "t40_ms": 50.0,
+ "active_s": 0.069,
+ "onset_s": 0.006,
+ "env_oct": 1.323,
+ "curve_db": 11.4,
+ "t40_ratio": 0.556,
+ "attack_ratio": 6.0,
+ "active_ratio": 0.69,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 90.0,
+ "active_s": 0.1
+ },
+ "phrase_z": 1.7,
+ "phrase_min": 1.03,
+ "margin": 0.18,
+ "wins": 0.0,
+ "best_confusion": "a knock on a door",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.073
+ ],
+ [
+ "Music",
+ 0.055
+ ],
+ [
+ "Speech",
+ 0.047
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.005,
+ "floor_db": -40.4,
+ "stereo_corr": 1.0,
+ "darken": 2.29,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -72.6,
+ "first_sample": 0.0,
+ "tail_db": -50.7,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/tap/bench/fs-707827_3.wav",
+ "length_s": 0.088,
+ "typ_z": -2.82,
+ "nn_cos": 0.723,
+ "tone_db": 13.49,
+ "tone_ratio": 2.09,
+ "n_outside": 8,
+ "outside": [
+ [
+ "39 Hz",
+ -46.2,
+ "below"
+ ],
+ [
+ "12.7 kHz",
+ -21.1,
+ "below"
+ ],
+ [
+ "16 kHz",
+ -20.1,
+ "below"
+ ]
+ ],
+ "attack_ms": 8.0,
+ "t40_ms": 55.0,
+ "active_s": 0.064,
+ "onset_s": 0.006,
+ "env_oct": 1.451,
+ "curve_db": 11.37,
+ "t40_ratio": 0.611,
+ "attack_ratio": 8.0,
+ "active_ratio": 0.64,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 90.0,
+ "active_s": 0.1
+ },
+ "phrase_z": 1.55,
+ "phrase_min": 0.82,
+ "margin": 0.51,
+ "wins": 0.0,
+ "best_confusion": "a knock on a door",
+ "as_top": [
+ [
+ "Music",
+ 0.091
+ ],
+ [
+ "Speech",
+ 0.077
+ ],
+ [
+ "Sound effect",
+ 0.068
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.005,
+ "floor_db": -43.0,
+ "stereo_corr": 1.0,
+ "darken": 0.94,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -87.6,
+ "first_sample": 0.0,
+ "tail_db": -53.0,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity_bench": {
+ "n": 6,
+ "cos_mean": 0.716,
+ "cos_max": 0.98,
+ "tone_min_db": 2.68,
+ "tone_mean_db": 14.56,
+ "near_duplicates": []
+ },
+ "diversity_forge": {
+ "n": 7,
+ "cos_mean": 0.713,
+ "cos_max": 0.904,
+ "tone_min_db": 3.1,
+ "tone_mean_db": 8.76,
+ "near_duplicates": []
+ },
+ "summary": {
+ "made": {},
+ "real_cross": {
+ "n": 7,
+ "typ_z": -1.17,
+ "tone_db": 9.35,
+ "env_oct": 1.59,
+ "curve_db": 13.67,
+ "phrase_z": 1.85,
+ "margin": 0.92,
+ "as_p": 0.01,
+ "tone_ratio": 0.56,
+ "t40_ratio": 1.64,
+ "wins": 0.0,
+ "floor_db": -43.6,
+ "stereo_corr": 1.0,
+ "darken": 1.31,
+ "as_names": 0.14,
+ "artifacts": 0,
+ "var_cos": 0.713
+ },
+ "real_loo": {
+ "n": 6,
+ "typ_z": 1.66,
+ "tone_db": 16.78,
+ "env_oct": 1.19,
+ "curve_db": 11.59,
+ "phrase_z": 1.62,
+ "margin": 0.35,
+ "as_p": 0.01,
+ "tone_ratio": 1.0,
+ "t40_ratio": 1.0,
+ "wins": 0.0,
+ "floor_db": -38.4,
+ "stereo_corr": 1.0,
+ "darken": 0.94,
+ "as_names": 0.17,
+ "artifacts": 0,
+ "var_cos": 0.716
+ }
+ }
+ },
+ "toggle": {
+ "n_bench": 4,
+ "n_forge": 4,
+ "note": "",
+ "bench_refs": [
+ "E:/earshot/out/refs/toggle/bench/fs-278205_1.wav",
+ "E:/earshot/out/refs/toggle/bench/mk-2585_1.wav",
+ "E:/earshot/out/refs/toggle/bench/fs-401803_1.wav",
+ "E:/earshot/out/refs/toggle/bench/fs-401803_2.wav"
+ ],
+ "made": [],
+ "diversity": {
+ "n": 0
+ },
+ "real_cross": [
+ {
+ "file": "E:/earshot/out/refs/toggle/forge/kit-ui_toggle_01.wav",
+ "length_s": 0.19,
+ "typ_z": -0.17,
+ "nn_cos": 0.768,
+ "tone_db": 8.41,
+ "tone_ratio": 1.35,
+ "n_outside": 10,
+ "outside": [
+ [
+ "39 Hz",
+ 16.4,
+ "above"
+ ],
+ [
+ "62 Hz",
+ 15.7,
+ "above"
+ ],
+ [
+ "79 Hz",
+ 14.4,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 75.0,
+ "active_s": 0.074,
+ "onset_s": 0.006,
+ "env_oct": 0.088,
+ "curve_db": 4.79,
+ "t40_ratio": 1.111,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.08,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 67.5,
+ "active_s": 0.069
+ },
+ "phrase_z": 2.58,
+ "phrase_min": 2.49,
+ "margin": 0.28,
+ "wins": 0.5,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.113
+ ],
+ [
+ "Slap, smack",
+ 0.085
+ ],
+ [
+ "Sound effect",
+ 0.052
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.001,
+ "floor_db": -62.9,
+ "stereo_corr": 0.76,
+ "darken": -0.08,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -84.6,
+ "first_sample": 0.0,
+ "tail_db": -57.6,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/toggle/forge/kit-ui_toggle_02.wav",
+ "length_s": 0.24,
+ "typ_z": -0.31,
+ "nn_cos": 0.751,
+ "tone_db": 13.13,
+ "tone_ratio": 2.1,
+ "n_outside": 13,
+ "outside": [
+ [
+ "79 Hz",
+ 28.8,
+ "above"
+ ],
+ [
+ "62 Hz",
+ 28.3,
+ "above"
+ ],
+ [
+ "50 Hz",
+ 27.7,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 125.0,
+ "active_s": 0.123,
+ "onset_s": 0.012,
+ "env_oct": 0.578,
+ "curve_db": 9.11,
+ "t40_ratio": 1.852,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.796,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 67.5,
+ "active_s": 0.069
+ },
+ "phrase_z": 2.63,
+ "phrase_min": 2.52,
+ "margin": -0.1,
+ "wins": 0.5,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Slap, smack",
+ 0.204
+ ],
+ [
+ "Cap gun",
+ 0.031
+ ],
+ [
+ "Gunshot, gunfire",
+ 0.03
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.0,
+ "floor_db": -51.9,
+ "stereo_corr": 1.0,
+ "darken": -0.14,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -102.8,
+ "first_sample": 0.0,
+ "tail_db": -56.1,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/toggle/forge/fs-364200_1.wav",
+ "length_s": 0.365,
+ "typ_z": -8.13,
+ "nn_cos": 0.409,
+ "tone_db": 23.8,
+ "tone_ratio": 3.81,
+ "n_outside": 24,
+ "outside": [
+ [
+ "12.7 kHz",
+ -45.6,
+ "below"
+ ],
+ [
+ "10.1 kHz",
+ -44.3,
+ "below"
+ ],
+ [
+ "16 kHz",
+ -42.4,
+ "below"
+ ]
+ ],
+ "attack_ms": 55.0,
+ "t40_ms": 180.0,
+ "active_s": 0.331,
+ "onset_s": 0.004,
+ "env_oct": 3.156,
+ "curve_db": 33.61,
+ "t40_ratio": 2.667,
+ "attack_ratio": 55.0,
+ "active_ratio": 4.832,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 67.5,
+ "active_s": 0.069
+ },
+ "phrase_z": 1.61,
+ "phrase_min": 1.1,
+ "margin": 0.06,
+ "wins": 0.0,
+ "best_confusion": "a door latch",
+ "as_top": [
+ [
+ "Door",
+ 0.263
+ ],
+ [
+ "Slam",
+ 0.123
+ ],
+ [
+ "Music",
+ 0.045
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.001,
+ "floor_db": -44.7,
+ "stereo_corr": 0.998,
+ "darken": -0.79,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -47.3,
+ "first_sample": 0.0,
+ "tail_db": -63.5,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/toggle/forge/fs-448393_1.wav",
+ "length_s": 0.45,
+ "typ_z": -5.38,
+ "nn_cos": 0.493,
+ "tone_db": 11.84,
+ "tone_ratio": 1.89,
+ "n_outside": 19,
+ "outside": [
+ [
+ "198 Hz",
+ 21.5,
+ "above"
+ ],
+ [
+ "62 Hz",
+ 19.6,
+ "above"
+ ],
+ [
+ "79 Hz",
+ 19.4,
+ "above"
+ ]
+ ],
+ "attack_ms": 160.0,
+ "t40_ms": 160.0,
+ "active_s": 0.319,
+ "onset_s": 0.066,
+ "env_oct": 3.595,
+ "curve_db": 28.45,
+ "t40_ratio": 2.37,
+ "attack_ratio": 160.0,
+ "active_ratio": 4.657,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 67.5,
+ "active_s": 0.069
+ },
+ "phrase_z": 0.72,
+ "phrase_min": 0.45,
+ "margin": -2.2,
+ "wins": 0.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Crack",
+ 0.642
+ ],
+ [
+ "Crushing",
+ 0.028
+ ],
+ [
+ "Breaking",
+ 0.026
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.0,
+ "floor_db": -44.9,
+ "stereo_corr": 1.0,
+ "darken": 0.18,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -100.2,
+ "first_sample": 0.0,
+ "tail_db": -66.2,
+ "lead_s": 0.004,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_loo": [
+ {
+ "file": "E:/earshot/out/refs/toggle/bench/fs-278205_1.wav",
+ "length_s": 0.125,
+ "typ_z": -5.22,
+ "nn_cos": 0.678,
+ "tone_db": 7.48,
+ "tone_ratio": 1.46,
+ "n_outside": 10,
+ "outside": [
+ [
+ "99 Hz",
+ -16.4,
+ "below"
+ ],
+ [
+ "125 Hz",
+ -14.9,
+ "below"
+ ],
+ [
+ "3.17 kHz",
+ -11.4,
+ "below"
+ ]
+ ],
+ "attack_ms": 2.0,
+ "t40_ms": 60.0,
+ "active_s": 0.063,
+ "onset_s": 0.032,
+ "env_oct": 0.518,
+ "curve_db": 15.45,
+ "t40_ratio": 0.8,
+ "attack_ratio": 2.0,
+ "active_ratio": 0.851,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 75.0,
+ "active_s": 0.074
+ },
+ "phrase_z": 1.1,
+ "phrase_min": 1.07,
+ "margin": -0.93,
+ "wins": 0.0,
+ "best_confusion": "a seatbelt buckle",
+ "as_top": [
+ [
+ "Ding",
+ 0.099
+ ],
+ [
+ "Coin (dropping)",
+ 0.076
+ ],
+ [
+ "Chink, clink",
+ 0.074
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -55.4,
+ "stereo_corr": 0.638,
+ "darken": -0.19,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -97.7,
+ "first_sample": 0.0,
+ "tail_db": -65.9,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/toggle/bench/mk-2585_1.wav",
+ "length_s": 0.1,
+ "typ_z": 1.33,
+ "nn_cos": 0.836,
+ "tone_db": 7.36,
+ "tone_ratio": 1.43,
+ "n_outside": 11,
+ "outside": [
+ [
+ "2.52 kHz",
+ 13.6,
+ "above"
+ ],
+ [
+ "10.1 kHz",
+ -13.2,
+ "below"
+ ],
+ [
+ "630 Hz",
+ 11.0,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 55.0,
+ "active_s": 0.054,
+ "onset_s": 0.001,
+ "env_oct": 0.301,
+ "curve_db": 5.9,
+ "t40_ratio": 0.733,
+ "attack_ratio": 1.0,
+ "active_ratio": 0.73,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 75.0,
+ "active_s": 0.074
+ },
+ "phrase_z": 1.99,
+ "phrase_min": 1.86,
+ "margin": 0.06,
+ "wins": 0.0,
+ "best_confusion": "a door latch",
+ "as_top": [
+ [
+ "Speech",
+ 0.27
+ ],
+ [
+ "Music",
+ 0.056
+ ],
+ [
+ "Sound effect",
+ 0.03
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -55.5,
+ "stereo_corr": 1.0,
+ "darken": 0.14,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -80.5,
+ "first_sample": 0.0,
+ "tail_db": -62.2,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/toggle/bench/fs-401803_1.wav",
+ "length_s": 0.1,
+ "typ_z": 5.98,
+ "nn_cos": 0.846,
+ "tone_db": 5.13,
+ "tone_ratio": 0.7,
+ "n_outside": 3,
+ "outside": [
+ [
+ "62 Hz",
+ 14.6,
+ "above"
+ ],
+ [
+ "315 Hz",
+ -7.7,
+ "below"
+ ],
+ [
+ "79 Hz",
+ 6.9,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 80.0,
+ "active_s": 0.08,
+ "onset_s": 0.005,
+ "env_oct": 0.253,
+ "curve_db": 3.93,
+ "t40_ratio": 1.333,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.27,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 60.0,
+ "active_s": 0.063
+ },
+ "phrase_z": 2.0,
+ "phrase_min": 1.92,
+ "margin": -0.2,
+ "wins": 0.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Coin (dropping)",
+ 0.209
+ ],
+ [
+ "Speech",
+ 0.128
+ ],
+ [
+ "Cutlery, silverware",
+ 0.077
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.01,
+ "floor_db": -40.4,
+ "stereo_corr": 0.322,
+ "darken": -0.21,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -61.6,
+ "first_sample": 0.0,
+ "tail_db": -52.6,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/toggle/bench/fs-401803_2.wav",
+ "length_s": 0.095,
+ "typ_z": 1.74,
+ "nn_cos": 0.846,
+ "tone_db": 4.67,
+ "tone_ratio": 0.64,
+ "n_outside": 1,
+ "outside": [
+ [
+ "2 kHz",
+ -13.5,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 75.0,
+ "active_s": 0.074,
+ "onset_s": 0.006,
+ "env_oct": 0.185,
+ "curve_db": 3.37,
+ "t40_ratio": 1.25,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.175,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 60.0,
+ "active_s": 0.063
+ },
+ "phrase_z": 1.89,
+ "phrase_min": 1.86,
+ "margin": -0.35,
+ "wins": 0.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Coin (dropping)",
+ 0.252
+ ],
+ [
+ "Speech",
+ 0.088
+ ],
+ [
+ "Single-lens reflex camera",
+ 0.067
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.004,
+ "floor_db": -42.8,
+ "stereo_corr": 0.213,
+ "darken": -0.18,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -60.4,
+ "first_sample": 0.0,
+ "tail_db": -49.7,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_rev": [
+ {
+ "file": "E:/earshot/out/refs/toggle/bench/fs-278205_1.wav",
+ "length_s": 0.125,
+ "typ_z": 0.39,
+ "nn_cos": 0.515,
+ "tone_db": 15.72,
+ "tone_ratio": 1.64,
+ "n_outside": 15,
+ "outside": [
+ [
+ "62 Hz",
+ -29.6,
+ "below"
+ ],
+ [
+ "31 Hz",
+ -26.0,
+ "below"
+ ],
+ [
+ "50 Hz",
+ -25.3,
+ "below"
+ ]
+ ],
+ "attack_ms": 2.0,
+ "t40_ms": 60.0,
+ "active_s": 0.063,
+ "onset_s": 0.032,
+ "env_oct": 2.289,
+ "curve_db": 15.11,
+ "t40_ratio": 0.421,
+ "attack_ratio": 0.071,
+ "active_ratio": 0.285,
+ "ref_median": {
+ "attack_ms": 28.0,
+ "t40_ms": 142.5,
+ "active_s": 0.221
+ },
+ "phrase_z": 1.1,
+ "phrase_min": 1.07,
+ "margin": -0.93,
+ "wins": 0.0,
+ "best_confusion": "a seatbelt buckle",
+ "as_top": [
+ [
+ "Ding",
+ 0.099
+ ],
+ [
+ "Coin (dropping)",
+ 0.076
+ ],
+ [
+ "Chink, clink",
+ 0.074
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -55.4,
+ "stereo_corr": 0.638,
+ "darken": -0.19,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -97.7,
+ "first_sample": 0.0,
+ "tail_db": -65.9,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/toggle/bench/mk-2585_1.wav",
+ "length_s": 0.1,
+ "typ_z": 2.87,
+ "nn_cos": 0.751,
+ "tone_db": 8.46,
+ "tone_ratio": 0.88,
+ "n_outside": 7,
+ "outside": [
+ [
+ "79 Hz",
+ -21.3,
+ "below"
+ ],
+ [
+ "62 Hz",
+ -18.4,
+ "below"
+ ],
+ [
+ "50 Hz",
+ -17.1,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 55.0,
+ "active_s": 0.054,
+ "onset_s": 0.001,
+ "env_oct": 2.738,
+ "curve_db": 15.55,
+ "t40_ratio": 0.386,
+ "attack_ratio": 0.036,
+ "active_ratio": 0.244,
+ "ref_median": {
+ "attack_ms": 28.0,
+ "t40_ms": 142.5,
+ "active_s": 0.221
+ },
+ "phrase_z": 1.99,
+ "phrase_min": 1.86,
+ "margin": 0.06,
+ "wins": 0.0,
+ "best_confusion": "a door latch",
+ "as_top": [
+ [
+ "Speech",
+ 0.27
+ ],
+ [
+ "Music",
+ 0.056
+ ],
+ [
+ "Sound effect",
+ 0.03
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -55.5,
+ "stereo_corr": 1.0,
+ "darken": 0.14,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -80.5,
+ "first_sample": 0.0,
+ "tail_db": -62.2,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/toggle/bench/fs-401803_1.wav",
+ "length_s": 0.1,
+ "typ_z": 2.4,
+ "nn_cos": 0.768,
+ "tone_db": 10.2,
+ "tone_ratio": 1.06,
+ "n_outside": 8,
+ "outside": [
+ [
+ "31 Hz",
+ -19.3,
+ "below"
+ ],
+ [
+ "39 Hz",
+ -18.1,
+ "below"
+ ],
+ [
+ "50 Hz",
+ -16.9,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 80.0,
+ "active_s": 0.08,
+ "onset_s": 0.005,
+ "env_oct": 2.369,
+ "curve_db": 14.45,
+ "t40_ratio": 0.561,
+ "attack_ratio": 0.036,
+ "active_ratio": 0.362,
+ "ref_median": {
+ "attack_ms": 28.0,
+ "t40_ms": 142.5,
+ "active_s": 0.221
+ },
+ "phrase_z": 2.0,
+ "phrase_min": 1.92,
+ "margin": -0.2,
+ "wins": 0.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Coin (dropping)",
+ 0.209
+ ],
+ [
+ "Speech",
+ 0.128
+ ],
+ [
+ "Cutlery, silverware",
+ 0.077
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.01,
+ "floor_db": -40.4,
+ "stereo_corr": 0.322,
+ "darken": -0.21,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -61.6,
+ "first_sample": 0.0,
+ "tail_db": -52.6,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/toggle/bench/fs-401803_2.wav",
+ "length_s": 0.095,
+ "typ_z": 0.84,
+ "nn_cos": 0.637,
+ "tone_db": 13.41,
+ "tone_ratio": 1.4,
+ "n_outside": 11,
+ "outside": [
+ [
+ "62 Hz",
+ -29.6,
+ "below"
+ ],
+ [
+ "39 Hz",
+ -25.0,
+ "below"
+ ],
+ [
+ "50 Hz",
+ -24.1,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 75.0,
+ "active_s": 0.074,
+ "onset_s": 0.006,
+ "env_oct": 2.437,
+ "curve_db": 14.52,
+ "t40_ratio": 0.526,
+ "attack_ratio": 0.036,
+ "active_ratio": 0.335,
+ "ref_median": {
+ "attack_ms": 28.0,
+ "t40_ms": 142.5,
+ "active_s": 0.221
+ },
+ "phrase_z": 1.89,
+ "phrase_min": 1.86,
+ "margin": -0.35,
+ "wins": 0.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Coin (dropping)",
+ 0.252
+ ],
+ [
+ "Speech",
+ 0.088
+ ],
+ [
+ "Single-lens reflex camera",
+ 0.067
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.004,
+ "floor_db": -42.8,
+ "stereo_corr": 0.213,
+ "darken": -0.18,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -60.4,
+ "first_sample": 0.0,
+ "tail_db": -49.7,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity_bench": {
+ "n": 4,
+ "cos_mean": 0.725,
+ "cos_max": 0.846,
+ "tone_min_db": 5.33,
+ "tone_mean_db": 7.21,
+ "near_duplicates": []
+ },
+ "diversity_forge": {
+ "n": 4,
+ "cos_mean": 0.477,
+ "cos_max": 0.724,
+ "tone_min_db": 9.53,
+ "tone_mean_db": 13.88,
+ "near_duplicates": []
+ },
+ "summary": {
+ "made": {},
+ "real_cross": {
+ "n": 4,
+ "typ_z": -2.84,
+ "tone_db": 12.48,
+ "env_oct": 1.87,
+ "curve_db": 18.78,
+ "phrase_z": 2.1,
+ "margin": -0.02,
+ "as_p": 0.0,
+ "tone_ratio": 2.0,
+ "t40_ratio": 2.11,
+ "wins": 0.25,
+ "floor_db": -48.4,
+ "stereo_corr": 1.0,
+ "darken": -0.11,
+ "as_names": 0.0,
+ "artifacts": 0,
+ "var_cos": 0.477
+ },
+ "real_loo": {
+ "n": 4,
+ "typ_z": 1.54,
+ "tone_db": 6.25,
+ "env_oct": 0.28,
+ "curve_db": 4.92,
+ "phrase_z": 1.94,
+ "margin": -0.28,
+ "as_p": 0.0,
+ "tone_ratio": 1.06,
+ "t40_ratio": 1.02,
+ "wins": 0.0,
+ "floor_db": -49.1,
+ "stereo_corr": 0.48,
+ "darken": -0.18,
+ "as_names": 0.0,
+ "artifacts": 0,
+ "var_cos": 0.725
+ }
+ }
+ },
+ "typing_key": {
+ "n_bench": 12,
+ "n_forge": 15,
+ "note": "",
+ "bench_refs": [
+ "E:/earshot/out/refs/typing_key/bench/fs-378083_1.wav",
+ "E:/earshot/out/refs/typing_key/bench/fs-378084_1.wav",
+ "E:/earshot/out/refs/typing_key/bench/fs-378085_1.wav",
+ "E:/earshot/out/refs/typing_key/bench/fs-570754_1.wav",
+ "E:/earshot/out/refs/typing_key/bench/fs-766639_1.wav",
+ "E:/earshot/out/refs/typing_key/bench/fs-766640_1.wav",
+ "E:/earshot/out/refs/typing_key/bench/fs-609534_1.wav",
+ "E:/earshot/out/refs/typing_key/bench/fs-609534_2.wav",
+ "E:/earshot/out/refs/typing_key/bench/fs-609534_3.wav",
+ "E:/earshot/out/refs/typing_key/bench/fs-515810_1.wav",
+ "E:/earshot/out/refs/typing_key/bench/fs-815614_1.wav",
+ "E:/earshot/out/refs/typing_key/bench/fs-815614_2.wav"
+ ],
+ "made": [],
+ "diversity": {
+ "n": 0
+ },
+ "real_cross": [
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/kit-kb_thock_01.wav",
+ "length_s": 0.254,
+ "typ_z": 1.34,
+ "nn_cos": 0.926,
+ "tone_db": 9.89,
+ "tone_ratio": 1.81,
+ "n_outside": 7,
+ "outside": [
+ [
+ "31 Hz",
+ -17.6,
+ "below"
+ ],
+ [
+ "630 Hz",
+ -17.3,
+ "below"
+ ],
+ [
+ "794 Hz",
+ -16.2,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 205.0,
+ "active_s": 0.218,
+ "onset_s": 0.012,
+ "env_oct": 1.033,
+ "curve_db": 15.21,
+ "t40_ratio": 1.708,
+ "attack_ratio": 0.2,
+ "active_ratio": 0.998,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 5.48,
+ "phrase_min": 5.14,
+ "margin": 3.64,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.306
+ ],
+ [
+ "Computer keyboard",
+ 0.076
+ ],
+ [
+ "Tap",
+ 0.058
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.076,
+ "floor_db": -40.0,
+ "stereo_corr": 0.668,
+ "darken": -0.39,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -97.7,
+ "first_sample": 0.0,
+ "tail_db": -53.1,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/kit-kb_thock_02.wav",
+ "length_s": 0.281,
+ "typ_z": 0.98,
+ "nn_cos": 0.905,
+ "tone_db": 10.24,
+ "tone_ratio": 1.87,
+ "n_outside": 5,
+ "outside": [
+ [
+ "630 Hz",
+ -23.9,
+ "below"
+ ],
+ [
+ "62 Hz",
+ 17.4,
+ "above"
+ ],
+ [
+ "99 Hz",
+ 15.3,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 175.0,
+ "active_s": 0.255,
+ "onset_s": 0.005,
+ "env_oct": 1.03,
+ "curve_db": 19.49,
+ "t40_ratio": 1.458,
+ "attack_ratio": 0.2,
+ "active_ratio": 1.167,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 5.18,
+ "phrase_min": 4.93,
+ "margin": 3.4,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Computer keyboard",
+ 0.481
+ ],
+ [
+ "Typing",
+ 0.285
+ ],
+ [
+ "Sound effect",
+ 0.05
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.481,
+ "floor_db": -38.9,
+ "stereo_corr": 0.656,
+ "darken": 0.13,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -95.8,
+ "first_sample": 0.0,
+ "tail_db": -61.0,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/kit-kb_thock_03.wav",
+ "length_s": 0.362,
+ "typ_z": 1.25,
+ "nn_cos": 0.927,
+ "tone_db": 9.38,
+ "tone_ratio": 1.71,
+ "n_outside": 4,
+ "outside": [
+ [
+ "630 Hz",
+ -17.6,
+ "below"
+ ],
+ [
+ "62 Hz",
+ 15.8,
+ "above"
+ ],
+ [
+ "500 Hz",
+ -14.0,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 185.0,
+ "active_s": 0.269,
+ "onset_s": 0.041,
+ "env_oct": 1.082,
+ "curve_db": 14.68,
+ "t40_ratio": 1.542,
+ "attack_ratio": 0.2,
+ "active_ratio": 1.231,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 5.42,
+ "phrase_min": 4.97,
+ "margin": 3.87,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Computer keyboard",
+ 0.528
+ ],
+ [
+ "Typing",
+ 0.355
+ ],
+ [
+ "Sound effect",
+ 0.076
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.528,
+ "floor_db": -51.2,
+ "stereo_corr": 0.591,
+ "darken": 0.01,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -96.7,
+ "first_sample": 0.0,
+ "tail_db": -59.9,
+ "lead_s": 0.005,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/kit-kb_thock_04.wav",
+ "length_s": 0.234,
+ "typ_z": 1.35,
+ "nn_cos": 0.908,
+ "tone_db": 7.72,
+ "tone_ratio": 1.41,
+ "n_outside": 2,
+ "outside": [
+ [
+ "62 Hz",
+ 14.9,
+ "above"
+ ],
+ [
+ "500 Hz",
+ -9.5,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 175.0,
+ "active_s": 0.173,
+ "onset_s": 0.042,
+ "env_oct": 1.068,
+ "curve_db": 9.55,
+ "t40_ratio": 1.458,
+ "attack_ratio": 0.2,
+ "active_ratio": 0.792,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 4.75,
+ "phrase_min": 4.42,
+ "margin": 2.83,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Computer keyboard",
+ 0.077
+ ],
+ [
+ "Sound effect",
+ 0.063
+ ],
+ [
+ "Typing",
+ 0.044
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.077,
+ "floor_db": -43.7,
+ "stereo_corr": 0.528,
+ "darken": -0.8,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -84.1,
+ "first_sample": 0.0,
+ "tail_db": -61.8,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/kit-kb_enter_01.wav",
+ "length_s": 0.4,
+ "typ_z": 1.26,
+ "nn_cos": 0.917,
+ "tone_db": 8.82,
+ "tone_ratio": 1.61,
+ "n_outside": 5,
+ "outside": [
+ [
+ "62 Hz",
+ 18.7,
+ "above"
+ ],
+ [
+ "5.04 kHz",
+ -16.7,
+ "below"
+ ],
+ [
+ "4 kHz",
+ -14.3,
+ "below"
+ ]
+ ],
+ "attack_ms": 3.0,
+ "t40_ms": 200.0,
+ "active_s": 0.274,
+ "onset_s": 0.016,
+ "env_oct": 0.6,
+ "curve_db": 19.85,
+ "t40_ratio": 1.667,
+ "attack_ratio": 0.6,
+ "active_ratio": 1.254,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 4.89,
+ "phrase_min": 4.43,
+ "margin": 3.05,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.167
+ ],
+ [
+ "Tap",
+ 0.064
+ ],
+ [
+ "Sound effect",
+ 0.062
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.05,
+ "floor_db": -53.9,
+ "stereo_corr": 0.76,
+ "darken": -0.83,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -89.8,
+ "first_sample": 0.0,
+ "tail_db": -80.9,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/kit-kb_space_01.wav",
+ "length_s": 0.33,
+ "typ_z": 0.01,
+ "nn_cos": 0.813,
+ "tone_db": 7.12,
+ "tone_ratio": 1.3,
+ "n_outside": 2,
+ "outside": [
+ [
+ "62 Hz",
+ 17.8,
+ "above"
+ ],
+ [
+ "79 Hz",
+ 14.0,
+ "above"
+ ]
+ ],
+ "attack_ms": 2.0,
+ "t40_ms": 160.0,
+ "active_s": 0.228,
+ "onset_s": 0.017,
+ "env_oct": 0.599,
+ "curve_db": 16.65,
+ "t40_ratio": 1.333,
+ "attack_ratio": 0.4,
+ "active_ratio": 1.043,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 5.25,
+ "phrase_min": 4.88,
+ "margin": 4.11,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.084
+ ],
+ [
+ "Crack",
+ 0.043
+ ],
+ [
+ "Creak",
+ 0.036
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.012,
+ "floor_db": -61.4,
+ "stereo_corr": 0.578,
+ "darken": 0.22,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -90.1,
+ "first_sample": 0.0,
+ "tail_db": -86.0,
+ "lead_s": 0.005,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/kit-kb_click_01.wav",
+ "length_s": 0.254,
+ "typ_z": -0.47,
+ "nn_cos": 0.812,
+ "tone_db": 14.97,
+ "tone_ratio": 2.73,
+ "n_outside": 7,
+ "outside": [
+ [
+ "62 Hz",
+ -37.7,
+ "below"
+ ],
+ [
+ "50 Hz",
+ -32.5,
+ "below"
+ ],
+ [
+ "79 Hz",
+ -29.5,
+ "below"
+ ]
+ ],
+ "attack_ms": 15.0,
+ "t40_ms": 120.0,
+ "active_s": 0.134,
+ "onset_s": 0.006,
+ "env_oct": 0.763,
+ "curve_db": 9.52,
+ "t40_ratio": 1.0,
+ "attack_ratio": 3.0,
+ "active_ratio": 0.613,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 4.4,
+ "phrase_min": 3.68,
+ "margin": 2.21,
+ "wins": 1.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Single-lens reflex camera",
+ 0.069
+ ],
+ [
+ "Coin (dropping)",
+ 0.038
+ ],
+ [
+ "Sound effect",
+ 0.03
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.008,
+ "floor_db": -59.7,
+ "stereo_corr": 1.0,
+ "darken": 0.66,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -84.8,
+ "first_sample": 0.0,
+ "tail_db": -74.1,
+ "lead_s": 0.005,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/kit-kb_click_02.wav",
+ "length_s": 0.284,
+ "typ_z": -0.29,
+ "nn_cos": 0.802,
+ "tone_db": 14.0,
+ "tone_ratio": 2.56,
+ "n_outside": 6,
+ "outside": [
+ [
+ "79 Hz",
+ -34.9,
+ "below"
+ ],
+ [
+ "62 Hz",
+ -30.8,
+ "below"
+ ],
+ [
+ "50 Hz",
+ -25.8,
+ "below"
+ ]
+ ],
+ "attack_ms": 3.0,
+ "t40_ms": 110.0,
+ "active_s": 0.109,
+ "onset_s": 0.011,
+ "env_oct": 0.622,
+ "curve_db": 9.78,
+ "t40_ratio": 0.917,
+ "attack_ratio": 0.6,
+ "active_ratio": 0.499,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 4.69,
+ "phrase_min": 4.18,
+ "margin": 1.99,
+ "wins": 1.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Single-lens reflex camera",
+ 0.069
+ ],
+ [
+ "Coin (dropping)",
+ 0.055
+ ],
+ [
+ "Sound effect",
+ 0.04
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.012,
+ "floor_db": -67.6,
+ "stereo_corr": 1.0,
+ "darken": 0.72,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -79.1,
+ "first_sample": 0.0,
+ "tail_db": -88.7,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/kit-kb_enter_02.wav",
+ "length_s": 0.18,
+ "typ_z": -0.43,
+ "nn_cos": 0.783,
+ "tone_db": 8.88,
+ "tone_ratio": 1.62,
+ "n_outside": 1,
+ "outside": [
+ [
+ "16 kHz",
+ 20.8,
+ "above"
+ ]
+ ],
+ "attack_ms": 10.0,
+ "t40_ms": 155.0,
+ "active_s": 0.164,
+ "onset_s": 0.006,
+ "env_oct": 0.594,
+ "curve_db": 9.68,
+ "t40_ratio": 1.292,
+ "attack_ratio": 2.0,
+ "active_ratio": 0.751,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 3.33,
+ "phrase_min": 3.15,
+ "margin": 1.24,
+ "wins": 1.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Typewriter",
+ 0.088
+ ],
+ [
+ "Computer keyboard",
+ 0.072
+ ],
+ [
+ "Speech",
+ 0.056
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.088,
+ "floor_db": -36.4,
+ "stereo_corr": 1.0,
+ "darken": -0.5,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -95.9,
+ "first_sample": 0.0,
+ "tail_db": -46.3,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/fs-561697_1.wav",
+ "length_s": 0.18,
+ "typ_z": 0.26,
+ "nn_cos": 0.845,
+ "tone_db": 15.76,
+ "tone_ratio": 2.88,
+ "n_outside": 8,
+ "outside": [
+ [
+ "50 Hz",
+ -35.2,
+ "below"
+ ],
+ [
+ "79 Hz",
+ -32.8,
+ "below"
+ ],
+ [
+ "62 Hz",
+ -30.3,
+ "below"
+ ]
+ ],
+ "attack_ms": 17.0,
+ "t40_ms": 140.0,
+ "active_s": 0.155,
+ "onset_s": 0.005,
+ "env_oct": 0.828,
+ "curve_db": 17.09,
+ "t40_ratio": 1.167,
+ "attack_ratio": 3.4,
+ "active_ratio": 0.709,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 4.18,
+ "phrase_min": 3.68,
+ "margin": 1.42,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Single-lens reflex camera",
+ 0.34
+ ],
+ [
+ "Camera",
+ 0.045
+ ],
+ [
+ "Sound effect",
+ 0.043
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.012,
+ "floor_db": -40.4,
+ "stereo_corr": 1.0,
+ "darken": 0.04,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -73.5,
+ "first_sample": 0.0,
+ "tail_db": -56.6,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/fs-561699_1.wav",
+ "length_s": 0.419,
+ "typ_z": -0.98,
+ "nn_cos": 0.717,
+ "tone_db": 16.54,
+ "tone_ratio": 3.02,
+ "n_outside": 9,
+ "outside": [
+ [
+ "62 Hz",
+ -34.6,
+ "below"
+ ],
+ [
+ "79 Hz",
+ -33.1,
+ "below"
+ ],
+ [
+ "99 Hz",
+ -31.4,
+ "below"
+ ]
+ ],
+ "attack_ms": 17.0,
+ "t40_ms": 80.0,
+ "active_s": 0.094,
+ "onset_s": 0.021,
+ "env_oct": 1.189,
+ "curve_db": 8.6,
+ "t40_ratio": 0.667,
+ "attack_ratio": 3.4,
+ "active_ratio": 0.43,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 2.81,
+ "phrase_min": 2.37,
+ "margin": -0.23,
+ "wins": 0.5,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.054
+ ],
+ [
+ "Single-lens reflex camera",
+ 0.048
+ ],
+ [
+ "Coin (dropping)",
+ 0.028
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -62.9,
+ "stereo_corr": 1.0,
+ "darken": -0.66,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -85.3,
+ "first_sample": 0.0,
+ "tail_db": -68.9,
+ "lead_s": 0.004,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/fs-709460_1.wav",
+ "length_s": 0.278,
+ "typ_z": -0.15,
+ "nn_cos": 0.841,
+ "tone_db": 4.88,
+ "tone_ratio": 0.89,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 35.0,
+ "t40_ms": 195.0,
+ "active_s": 0.227,
+ "onset_s": 0.033,
+ "env_oct": 1.188,
+ "curve_db": 13.31,
+ "t40_ratio": 1.625,
+ "attack_ratio": 7.0,
+ "active_ratio": 1.039,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 3.96,
+ "phrase_min": 3.61,
+ "margin": 1.85,
+ "wins": 1.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Computer keyboard",
+ 0.264
+ ],
+ [
+ "Typing",
+ 0.17
+ ],
+ [
+ "Typewriter",
+ 0.077
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.264,
+ "floor_db": -40.0,
+ "stereo_corr": 0.603,
+ "darken": 0.7,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -92.1,
+ "first_sample": 0.0,
+ "tail_db": -47.3,
+ "lead_s": 0.004,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/fs-709460_2.wav",
+ "length_s": 0.11,
+ "typ_z": -0.15,
+ "nn_cos": 0.782,
+ "tone_db": 8.13,
+ "tone_ratio": 1.48,
+ "n_outside": 1,
+ "outside": [
+ [
+ "31 Hz",
+ -26.4,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 85.0,
+ "active_s": 0.085,
+ "onset_s": 0.005,
+ "env_oct": 1.394,
+ "curve_db": 12.2,
+ "t40_ratio": 0.708,
+ "attack_ratio": 0.2,
+ "active_ratio": 0.389,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 3.74,
+ "phrase_min": 3.42,
+ "margin": 2.0,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Music",
+ 0.143
+ ],
+ [
+ "Speech",
+ 0.129
+ ],
+ [
+ "Sound effect",
+ 0.045
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.001,
+ "floor_db": -44.2,
+ "stereo_corr": 0.675,
+ "darken": -0.12,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -96.2,
+ "first_sample": 0.0,
+ "tail_db": -50.8,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/fs-665075_1.wav",
+ "length_s": 0.198,
+ "typ_z": 0.97,
+ "nn_cos": 0.875,
+ "tone_db": 5.09,
+ "tone_ratio": 0.93,
+ "n_outside": 1,
+ "outside": [
+ [
+ "5.04 kHz",
+ -10.9,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 150.0,
+ "active_s": 0.166,
+ "onset_s": 0.014,
+ "env_oct": 1.013,
+ "curve_db": 9.28,
+ "t40_ratio": 1.25,
+ "attack_ratio": 0.2,
+ "active_ratio": 0.76,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 4.88,
+ "phrase_min": 4.6,
+ "margin": 3.11,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Computer keyboard",
+ 0.514
+ ],
+ [
+ "Typing",
+ 0.359
+ ],
+ [
+ "Speech",
+ 0.039
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.514,
+ "floor_db": -39.4,
+ "stereo_corr": 1.0,
+ "darken": 1.04,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -57.0,
+ "first_sample": 0.0,
+ "tail_db": -58.0,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/fs-665075_2.wav",
+ "length_s": 0.225,
+ "typ_z": 1.35,
+ "nn_cos": 0.903,
+ "tone_db": 7.38,
+ "tone_ratio": 1.35,
+ "n_outside": 1,
+ "outside": [
+ [
+ "50 Hz",
+ -23.0,
+ "below"
+ ]
+ ],
+ "attack_ms": 8.0,
+ "t40_ms": 130.0,
+ "active_s": 0.135,
+ "onset_s": 0.065,
+ "env_oct": 0.496,
+ "curve_db": 6.73,
+ "t40_ratio": 1.083,
+ "attack_ratio": 1.6,
+ "active_ratio": 0.618,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 4.95,
+ "phrase_min": 4.61,
+ "margin": 3.48,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Computer keyboard",
+ 0.687
+ ],
+ [
+ "Typing",
+ 0.517
+ ],
+ [
+ "Sound effect",
+ 0.04
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.687,
+ "floor_db": -42.5,
+ "stereo_corr": 1.0,
+ "darken": 0.45,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -57.0,
+ "first_sample": 0.0,
+ "tail_db": -54.5,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_loo": [
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-378083_1.wav",
+ "length_s": 0.17,
+ "typ_z": 0.15,
+ "nn_cos": 0.822,
+ "tone_db": 4.18,
+ "tone_ratio": 0.73,
+ "n_outside": 1,
+ "outside": [
+ [
+ "1 kHz",
+ -9.3,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 125.0,
+ "active_s": 0.125,
+ "onset_s": 0.035,
+ "env_oct": 1.321,
+ "curve_db": 7.32,
+ "t40_ratio": 1.087,
+ "attack_ratio": 0.143,
+ "active_ratio": 0.488,
+ "ref_median": {
+ "attack_ms": 7.0,
+ "t40_ms": 115.0,
+ "active_s": 0.256
+ },
+ "phrase_z": 3.24,
+ "phrase_min": 2.99,
+ "margin": 0.34,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.151
+ ],
+ [
+ "Computer keyboard",
+ 0.105
+ ],
+ [
+ "Typing",
+ 0.083
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.105,
+ "floor_db": -30.2,
+ "stereo_corr": 0.996,
+ "darken": 0.51,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -76.2,
+ "first_sample": 0.0,
+ "tail_db": -41.6,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-378084_1.wav",
+ "length_s": 0.367,
+ "typ_z": 0.88,
+ "nn_cos": 0.919,
+ "tone_db": 4.16,
+ "tone_ratio": 0.73,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 3.0,
+ "t40_ms": 180.0,
+ "active_s": 0.262,
+ "onset_s": 0.098,
+ "env_oct": 0.801,
+ "curve_db": 17.01,
+ "t40_ratio": 1.565,
+ "attack_ratio": 0.429,
+ "active_ratio": 1.448,
+ "ref_median": {
+ "attack_ms": 7.0,
+ "t40_ms": 115.0,
+ "active_s": 0.181
+ },
+ "phrase_z": 4.55,
+ "phrase_min": 4.29,
+ "margin": 2.2,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.445
+ ],
+ [
+ "Clapping",
+ 0.134
+ ],
+ [
+ "Speech",
+ 0.036
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.01,
+ "floor_db": -39.5,
+ "stereo_corr": 0.997,
+ "darken": 0.87,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -78.0,
+ "first_sample": 0.0,
+ "tail_db": -44.7,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-378085_1.wav",
+ "length_s": 0.16,
+ "typ_z": 0.45,
+ "nn_cos": 0.822,
+ "tone_db": 5.25,
+ "tone_ratio": 0.92,
+ "n_outside": 1,
+ "outside": [
+ [
+ "31 Hz",
+ -12.6,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 75.0,
+ "active_s": 0.126,
+ "onset_s": 0.009,
+ "env_oct": 1.522,
+ "curve_db": 10.62,
+ "t40_ratio": 0.6,
+ "attack_ratio": 0.143,
+ "active_ratio": 0.492,
+ "ref_median": {
+ "attack_ms": 7.0,
+ "t40_ms": 125.0,
+ "active_s": 0.256
+ },
+ "phrase_z": 3.83,
+ "phrase_min": 3.49,
+ "margin": 1.16,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.111
+ ],
+ [
+ "Slap, smack",
+ 0.061
+ ],
+ [
+ "Sound effect",
+ 0.035
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.001,
+ "floor_db": -49.6,
+ "stereo_corr": 0.999,
+ "darken": 0.1,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -79.3,
+ "first_sample": 0.0,
+ "tail_db": -60.5,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-570754_1.wav",
+ "length_s": 0.151,
+ "typ_z": 0.07,
+ "nn_cos": 0.822,
+ "tone_db": 5.26,
+ "tone_ratio": 0.92,
+ "n_outside": 2,
+ "outside": [
+ [
+ "2 kHz",
+ -9.7,
+ "below"
+ ],
+ [
+ "5.04 kHz",
+ 6.4,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 85.0,
+ "active_s": 0.085,
+ "onset_s": 0.045,
+ "env_oct": 1.651,
+ "curve_db": 9.86,
+ "t40_ratio": 0.68,
+ "attack_ratio": 0.143,
+ "active_ratio": 0.332,
+ "ref_median": {
+ "attack_ms": 7.0,
+ "t40_ms": 125.0,
+ "active_s": 0.256
+ },
+ "phrase_z": 3.3,
+ "phrase_min": 2.83,
+ "margin": 1.13,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.176
+ ],
+ [
+ "Coin (dropping)",
+ 0.151
+ ],
+ [
+ "Computer keyboard",
+ 0.049
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.049,
+ "floor_db": -47.8,
+ "stereo_corr": 1.0,
+ "darken": 0.08,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -72.9,
+ "first_sample": 0.0,
+ "tail_db": -56.8,
+ "lead_s": 0.014,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-766639_1.wav",
+ "length_s": 0.104,
+ "typ_z": 0.2,
+ "nn_cos": 0.815,
+ "tone_db": 9.0,
+ "tone_ratio": 1.71,
+ "n_outside": 2,
+ "outside": [
+ [
+ "315 Hz",
+ -26.1,
+ "below"
+ ],
+ [
+ "397 Hz",
+ -25.3,
+ "below"
+ ]
+ ],
+ "attack_ms": 21.0,
+ "t40_ms": 65.0,
+ "active_s": 0.083,
+ "onset_s": 0.002,
+ "env_oct": 1.792,
+ "curve_db": 14.07,
+ "t40_ratio": 0.52,
+ "attack_ratio": 7.0,
+ "active_ratio": 0.324,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 125.0,
+ "active_s": 0.256
+ },
+ "phrase_z": 3.58,
+ "phrase_min": 3.13,
+ "margin": 1.34,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Speech",
+ 0.145
+ ],
+ [
+ "Computer keyboard",
+ 0.077
+ ],
+ [
+ "Finger snapping",
+ 0.069
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.077,
+ "floor_db": -41.2,
+ "stereo_corr": 1.0,
+ "darken": 0.25,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -80.7,
+ "first_sample": 0.0,
+ "tail_db": -50.9,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-766640_1.wav",
+ "length_s": 0.194,
+ "typ_z": 0.11,
+ "nn_cos": 0.815,
+ "tone_db": 10.44,
+ "tone_ratio": 1.98,
+ "n_outside": 8,
+ "outside": [
+ [
+ "198 Hz",
+ -23.8,
+ "below"
+ ],
+ [
+ "157 Hz",
+ -19.8,
+ "below"
+ ],
+ [
+ "125 Hz",
+ -18.1,
+ "below"
+ ]
+ ],
+ "attack_ms": 136.0,
+ "t40_ms": 45.0,
+ "active_s": 0.181,
+ "onset_s": 0.009,
+ "env_oct": 2.492,
+ "curve_db": 19.85,
+ "t40_ratio": 0.36,
+ "attack_ratio": 45.333,
+ "active_ratio": 0.707,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 125.0,
+ "active_s": 0.256
+ },
+ "phrase_z": 3.23,
+ "phrase_min": 2.81,
+ "margin": 0.8,
+ "wins": 1.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.053
+ ],
+ [
+ "Scrape",
+ 0.032
+ ],
+ [
+ "Speech",
+ 0.03
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.0,
+ "floor_db": -35.4,
+ "stereo_corr": 1.0,
+ "darken": -0.67,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -82.5,
+ "first_sample": 0.0,
+ "tail_db": -37.6,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-609534_1.wav",
+ "length_s": 0.348,
+ "typ_z": 0.93,
+ "nn_cos": 0.919,
+ "tone_db": 5.69,
+ "tone_ratio": 1.08,
+ "n_outside": 1,
+ "outside": [
+ [
+ "79 Hz",
+ 12.5,
+ "above"
+ ]
+ ],
+ "attack_ms": 113.0,
+ "t40_ms": 145.0,
+ "active_s": 0.256,
+ "onset_s": 0.074,
+ "env_oct": 2.023,
+ "curve_db": 16.39,
+ "t40_ratio": 1.261,
+ "attack_ratio": 37.667,
+ "active_ratio": 1.414,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 115.0,
+ "active_s": 0.181
+ },
+ "phrase_z": 5.15,
+ "phrase_min": 4.76,
+ "margin": 3.08,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.123
+ ],
+ [
+ "Clapping",
+ 0.088
+ ],
+ [
+ "Ping",
+ 0.041
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.013,
+ "floor_db": -67.5,
+ "stereo_corr": 0.998,
+ "darken": -0.44,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -82.4,
+ "first_sample": 0.0,
+ "tail_db": -51.5,
+ "lead_s": 0.072,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-609534_2.wav",
+ "length_s": 0.442,
+ "typ_z": 0.89,
+ "nn_cos": 0.874,
+ "tone_db": 4.91,
+ "tone_ratio": 0.86,
+ "n_outside": 1,
+ "outside": [
+ [
+ "630 Hz",
+ -11.1,
+ "below"
+ ]
+ ],
+ "attack_ms": 99.0,
+ "t40_ms": 320.0,
+ "active_s": 0.418,
+ "onset_s": 0.007,
+ "env_oct": 2.576,
+ "curve_db": 26.12,
+ "t40_ratio": 2.783,
+ "attack_ratio": 33.0,
+ "active_ratio": 2.309,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 115.0,
+ "active_s": 0.181
+ },
+ "phrase_z": 4.65,
+ "phrase_min": 4.22,
+ "margin": 2.62,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Coin (dropping)",
+ 0.066
+ ],
+ [
+ "Sound effect",
+ 0.059
+ ],
+ [
+ "Single-lens reflex camera",
+ 0.04
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.014,
+ "floor_db": -36.7,
+ "stereo_corr": 0.999,
+ "darken": -2.83,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -83.3,
+ "first_sample": 0.0,
+ "tail_db": -52.1,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-609534_3.wav",
+ "length_s": 0.301,
+ "typ_z": -0.03,
+ "nn_cos": 0.809,
+ "tone_db": 4.86,
+ "tone_ratio": 0.85,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 1.0,
+ "t40_ms": 260.0,
+ "active_s": 0.269,
+ "onset_s": 0.006,
+ "env_oct": 1.519,
+ "curve_db": 20.05,
+ "t40_ratio": 2.261,
+ "attack_ratio": 0.143,
+ "active_ratio": 1.486,
+ "ref_median": {
+ "attack_ms": 7.0,
+ "t40_ms": 115.0,
+ "active_s": 0.181
+ },
+ "phrase_z": 4.84,
+ "phrase_min": 4.36,
+ "margin": 3.09,
+ "wins": 1.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Crack",
+ 0.126
+ ],
+ [
+ "Breaking",
+ 0.059
+ ],
+ [
+ "Sound effect",
+ 0.038
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.001,
+ "floor_db": -39.8,
+ "stereo_corr": 0.999,
+ "darken": 0.16,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -82.6,
+ "first_sample": 0.0,
+ "tail_db": -58.6,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-515810_1.wav",
+ "length_s": 0.319,
+ "typ_z": -12.68,
+ "nn_cos": 0.514,
+ "tone_db": 9.2,
+ "tone_ratio": 1.75,
+ "n_outside": 6,
+ "outside": [
+ [
+ "250 Hz",
+ 21.5,
+ "above"
+ ],
+ [
+ "12.7 kHz",
+ 20.9,
+ "above"
+ ],
+ [
+ "315 Hz",
+ 15.7,
+ "above"
+ ]
+ ],
+ "attack_ms": 7.0,
+ "t40_ms": 105.0,
+ "active_s": 0.288,
+ "onset_s": 0.012,
+ "env_oct": 0.715,
+ "curve_db": 23.77,
+ "t40_ratio": 0.84,
+ "attack_ratio": 2.333,
+ "active_ratio": 1.591,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 125.0,
+ "active_s": 0.181
+ },
+ "phrase_z": 2.32,
+ "phrase_min": 2.06,
+ "margin": 0.55,
+ "wins": 0.5,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.149
+ ],
+ [
+ "Speech",
+ 0.123
+ ],
+ [
+ "Computer keyboard",
+ 0.036
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.036,
+ "floor_db": -40.2,
+ "stereo_corr": 0.994,
+ "darken": -1.14,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -74.7,
+ "first_sample": 0.0,
+ "tail_db": -49.3,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-815614_1.wav",
+ "length_s": 0.269,
+ "typ_z": 0.25,
+ "nn_cos": 0.867,
+ "tone_db": 8.69,
+ "tone_ratio": 1.65,
+ "n_outside": 2,
+ "outside": [
+ [
+ "125 Hz",
+ 18.5,
+ "above"
+ ],
+ [
+ "99 Hz",
+ 12.3,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 260.0,
+ "active_s": 0.256,
+ "onset_s": 0.009,
+ "env_oct": 1.495,
+ "curve_db": 17.78,
+ "t40_ratio": 2.261,
+ "attack_ratio": 0.143,
+ "active_ratio": 1.414,
+ "ref_median": {
+ "attack_ms": 7.0,
+ "t40_ms": 115.0,
+ "active_s": 0.181
+ },
+ "phrase_z": 3.47,
+ "phrase_min": 3.2,
+ "margin": 1.72,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Tap",
+ 0.119
+ ],
+ [
+ "Music",
+ 0.116
+ ],
+ [
+ "Sound effect",
+ 0.088
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.016,
+ "floor_db": -27.3,
+ "stereo_corr": 1.0,
+ "darken": -0.09,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -70.5,
+ "first_sample": 0.0,
+ "tail_db": -31.0,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-815614_2.wav",
+ "length_s": 0.181,
+ "typ_z": 0.52,
+ "nn_cos": 0.867,
+ "tone_db": 8.02,
+ "tone_ratio": 1.52,
+ "n_outside": 1,
+ "outside": [
+ [
+ "794 Hz",
+ -6.8,
+ "below"
+ ]
+ ],
+ "attack_ms": 7.0,
+ "t40_ms": 115.0,
+ "active_s": 0.135,
+ "onset_s": 0.04,
+ "env_oct": 0.755,
+ "curve_db": 6.9,
+ "t40_ratio": 0.92,
+ "attack_ratio": 2.333,
+ "active_ratio": 0.527,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 125.0,
+ "active_s": 0.256
+ },
+ "phrase_z": 3.93,
+ "phrase_min": 3.57,
+ "margin": 2.23,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Computer keyboard",
+ 0.094
+ ],
+ [
+ "Speech",
+ 0.065
+ ],
+ [
+ "Tap",
+ 0.057
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.094,
+ "floor_db": -37.6,
+ "stereo_corr": 1.0,
+ "darken": 1.32,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -74.7,
+ "first_sample": 0.0,
+ "tail_db": -41.8,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_rev": [
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-378083_1.wav",
+ "length_s": 0.17,
+ "typ_z": -1.72,
+ "nn_cos": 0.748,
+ "tone_db": 6.31,
+ "tone_ratio": 0.74,
+ "n_outside": 1,
+ "outside": [
+ [
+ "10.1 kHz",
+ -9.8,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 125.0,
+ "active_s": 0.125,
+ "onset_s": 0.035,
+ "env_oct": 0.768,
+ "curve_db": 12.15,
+ "t40_ratio": 0.806,
+ "attack_ratio": 0.333,
+ "active_ratio": 0.753,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 155.0,
+ "active_s": 0.166
+ },
+ "phrase_z": 3.24,
+ "phrase_min": 2.99,
+ "margin": 0.34,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.151
+ ],
+ [
+ "Computer keyboard",
+ 0.105
+ ],
+ [
+ "Typing",
+ 0.083
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.105,
+ "floor_db": -30.2,
+ "stereo_corr": 0.996,
+ "darken": 0.51,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -76.2,
+ "first_sample": 0.0,
+ "tail_db": -41.6,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-378084_1.wav",
+ "length_s": 0.367,
+ "typ_z": 0.49,
+ "nn_cos": 0.892,
+ "tone_db": 8.06,
+ "tone_ratio": 0.95,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 3.0,
+ "t40_ms": 180.0,
+ "active_s": 0.262,
+ "onset_s": 0.098,
+ "env_oct": 0.291,
+ "curve_db": 16.7,
+ "t40_ratio": 1.161,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.578,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 155.0,
+ "active_s": 0.166
+ },
+ "phrase_z": 4.55,
+ "phrase_min": 4.29,
+ "margin": 2.2,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.445
+ ],
+ [
+ "Clapping",
+ 0.134
+ ],
+ [
+ "Speech",
+ 0.036
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.01,
+ "floor_db": -39.5,
+ "stereo_corr": 0.997,
+ "darken": 0.87,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -78.0,
+ "first_sample": 0.0,
+ "tail_db": -44.7,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-378085_1.wav",
+ "length_s": 0.16,
+ "typ_z": -1.31,
+ "nn_cos": 0.789,
+ "tone_db": 5.85,
+ "tone_ratio": 0.69,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 1.0,
+ "t40_ms": 75.0,
+ "active_s": 0.126,
+ "onset_s": 0.009,
+ "env_oct": 1.01,
+ "curve_db": 10.48,
+ "t40_ratio": 0.484,
+ "attack_ratio": 0.333,
+ "active_ratio": 0.759,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 155.0,
+ "active_s": 0.166
+ },
+ "phrase_z": 3.83,
+ "phrase_min": 3.49,
+ "margin": 1.16,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.111
+ ],
+ [
+ "Slap, smack",
+ 0.061
+ ],
+ [
+ "Sound effect",
+ 0.035
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.001,
+ "floor_db": -49.6,
+ "stereo_corr": 0.999,
+ "darken": 0.1,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -79.3,
+ "first_sample": 0.0,
+ "tail_db": -60.5,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-570754_1.wav",
+ "length_s": 0.151,
+ "typ_z": -1.82,
+ "nn_cos": 0.75,
+ "tone_db": 8.34,
+ "tone_ratio": 0.98,
+ "n_outside": 1,
+ "outside": [
+ [
+ "31 Hz",
+ 25.9,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 85.0,
+ "active_s": 0.085,
+ "onset_s": 0.045,
+ "env_oct": 1.139,
+ "curve_db": 14.5,
+ "t40_ratio": 0.548,
+ "attack_ratio": 0.333,
+ "active_ratio": 0.512,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 155.0,
+ "active_s": 0.166
+ },
+ "phrase_z": 3.3,
+ "phrase_min": 2.83,
+ "margin": 1.13,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.176
+ ],
+ [
+ "Coin (dropping)",
+ 0.151
+ ],
+ [
+ "Computer keyboard",
+ 0.049
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.049,
+ "floor_db": -47.8,
+ "stereo_corr": 1.0,
+ "darken": 0.08,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -72.9,
+ "first_sample": 0.0,
+ "tail_db": -56.8,
+ "lead_s": 0.014,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-766639_1.wav",
+ "length_s": 0.104,
+ "typ_z": -1.14,
+ "nn_cos": 0.789,
+ "tone_db": 8.4,
+ "tone_ratio": 0.99,
+ "n_outside": 3,
+ "outside": [
+ [
+ "315 Hz",
+ -21.5,
+ "below"
+ ],
+ [
+ "397 Hz",
+ -21.1,
+ "below"
+ ],
+ [
+ "630 Hz",
+ 12.9,
+ "above"
+ ]
+ ],
+ "attack_ms": 21.0,
+ "t40_ms": 65.0,
+ "active_s": 0.083,
+ "onset_s": 0.002,
+ "env_oct": 1.687,
+ "curve_db": 17.63,
+ "t40_ratio": 0.419,
+ "attack_ratio": 7.0,
+ "active_ratio": 0.5,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 155.0,
+ "active_s": 0.166
+ },
+ "phrase_z": 3.58,
+ "phrase_min": 3.13,
+ "margin": 1.34,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Speech",
+ 0.145
+ ],
+ [
+ "Computer keyboard",
+ 0.077
+ ],
+ [
+ "Finger snapping",
+ 0.069
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.077,
+ "floor_db": -41.2,
+ "stereo_corr": 1.0,
+ "darken": 0.25,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -80.7,
+ "first_sample": 0.0,
+ "tail_db": -50.9,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-766640_1.wav",
+ "length_s": 0.194,
+ "typ_z": -1.32,
+ "nn_cos": 0.771,
+ "tone_db": 10.12,
+ "tone_ratio": 1.19,
+ "n_outside": 2,
+ "outside": [
+ [
+ "198 Hz",
+ -18.9,
+ "below"
+ ],
+ [
+ "1 kHz",
+ 9.6,
+ "above"
+ ]
+ ],
+ "attack_ms": 136.0,
+ "t40_ms": 45.0,
+ "active_s": 0.181,
+ "onset_s": 0.009,
+ "env_oct": 2.471,
+ "curve_db": 24.3,
+ "t40_ratio": 0.29,
+ "attack_ratio": 45.333,
+ "active_ratio": 1.09,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 155.0,
+ "active_s": 0.166
+ },
+ "phrase_z": 3.23,
+ "phrase_min": 2.81,
+ "margin": 0.8,
+ "wins": 1.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.053
+ ],
+ [
+ "Scrape",
+ 0.032
+ ],
+ [
+ "Speech",
+ 0.03
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.0,
+ "floor_db": -35.4,
+ "stereo_corr": 1.0,
+ "darken": -0.67,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -82.5,
+ "first_sample": 0.0,
+ "tail_db": -37.6,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-609534_1.wav",
+ "length_s": 0.348,
+ "typ_z": 0.99,
+ "nn_cos": 0.927,
+ "tone_db": 5.06,
+ "tone_ratio": 0.59,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 113.0,
+ "t40_ms": 145.0,
+ "active_s": 0.256,
+ "onset_s": 0.074,
+ "env_oct": 1.985,
+ "curve_db": 19.66,
+ "t40_ratio": 0.935,
+ "attack_ratio": 37.667,
+ "active_ratio": 1.542,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 155.0,
+ "active_s": 0.166
+ },
+ "phrase_z": 5.15,
+ "phrase_min": 4.76,
+ "margin": 3.08,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.123
+ ],
+ [
+ "Clapping",
+ 0.088
+ ],
+ [
+ "Ping",
+ 0.041
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.013,
+ "floor_db": -67.5,
+ "stereo_corr": 0.998,
+ "darken": -0.44,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -82.4,
+ "first_sample": 0.0,
+ "tail_db": -51.5,
+ "lead_s": 0.072,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-609534_2.wav",
+ "length_s": 0.442,
+ "typ_z": 0.51,
+ "nn_cos": 0.903,
+ "tone_db": 6.7,
+ "tone_ratio": 0.79,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 99.0,
+ "t40_ms": 320.0,
+ "active_s": 0.418,
+ "onset_s": 0.007,
+ "env_oct": 2.474,
+ "curve_db": 26.86,
+ "t40_ratio": 2.065,
+ "attack_ratio": 33.0,
+ "active_ratio": 2.518,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 155.0,
+ "active_s": 0.166
+ },
+ "phrase_z": 4.65,
+ "phrase_min": 4.22,
+ "margin": 2.62,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Coin (dropping)",
+ 0.066
+ ],
+ [
+ "Sound effect",
+ 0.059
+ ],
+ [
+ "Single-lens reflex camera",
+ 0.04
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.014,
+ "floor_db": -36.7,
+ "stereo_corr": 0.999,
+ "darken": -2.83,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -83.3,
+ "first_sample": 0.0,
+ "tail_db": -52.1,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-609534_3.wav",
+ "length_s": 0.301,
+ "typ_z": -0.83,
+ "nn_cos": 0.823,
+ "tone_db": 5.61,
+ "tone_ratio": 0.66,
+ "n_outside": 1,
+ "outside": [
+ [
+ "1 kHz",
+ 11.3,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 260.0,
+ "active_s": 0.269,
+ "onset_s": 0.006,
+ "env_oct": 1.009,
+ "curve_db": 15.71,
+ "t40_ratio": 1.677,
+ "attack_ratio": 0.333,
+ "active_ratio": 1.62,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 155.0,
+ "active_s": 0.166
+ },
+ "phrase_z": 4.84,
+ "phrase_min": 4.36,
+ "margin": 3.09,
+ "wins": 1.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Crack",
+ 0.126
+ ],
+ [
+ "Breaking",
+ 0.059
+ ],
+ [
+ "Sound effect",
+ 0.038
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.001,
+ "floor_db": -39.8,
+ "stereo_corr": 0.999,
+ "darken": 0.16,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -82.6,
+ "first_sample": 0.0,
+ "tail_db": -58.6,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-515810_1.wav",
+ "length_s": 0.319,
+ "typ_z": -4.1,
+ "nn_cos": 0.583,
+ "tone_db": 11.19,
+ "tone_ratio": 1.32,
+ "n_outside": 5,
+ "outside": [
+ [
+ "250 Hz",
+ 27.4,
+ "above"
+ ],
+ [
+ "397 Hz",
+ 21.6,
+ "above"
+ ],
+ [
+ "315 Hz",
+ 20.3,
+ "above"
+ ]
+ ],
+ "attack_ms": 7.0,
+ "t40_ms": 105.0,
+ "active_s": 0.288,
+ "onset_s": 0.012,
+ "env_oct": 0.86,
+ "curve_db": 22.11,
+ "t40_ratio": 0.677,
+ "attack_ratio": 2.333,
+ "active_ratio": 1.735,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 155.0,
+ "active_s": 0.166
+ },
+ "phrase_z": 2.32,
+ "phrase_min": 2.06,
+ "margin": 0.55,
+ "wins": 0.5,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.149
+ ],
+ [
+ "Speech",
+ 0.123
+ ],
+ [
+ "Computer keyboard",
+ 0.036
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.036,
+ "floor_db": -40.2,
+ "stereo_corr": 0.994,
+ "darken": -1.14,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -74.7,
+ "first_sample": 0.0,
+ "tail_db": -49.3,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-815614_1.wav",
+ "length_s": 0.269,
+ "typ_z": -1.45,
+ "nn_cos": 0.764,
+ "tone_db": 11.51,
+ "tone_ratio": 1.35,
+ "n_outside": 5,
+ "outside": [
+ [
+ "198 Hz",
+ 22.6,
+ "above"
+ ],
+ [
+ "397 Hz",
+ 19.8,
+ "above"
+ ],
+ [
+ "157 Hz",
+ 19.7,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 260.0,
+ "active_s": 0.256,
+ "onset_s": 0.009,
+ "env_oct": 0.985,
+ "curve_db": 20.98,
+ "t40_ratio": 1.677,
+ "attack_ratio": 0.333,
+ "active_ratio": 1.542,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 155.0,
+ "active_s": 0.166
+ },
+ "phrase_z": 3.47,
+ "phrase_min": 3.2,
+ "margin": 1.72,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Tap",
+ 0.119
+ ],
+ [
+ "Music",
+ 0.116
+ ],
+ [
+ "Sound effect",
+ 0.088
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.016,
+ "floor_db": -27.3,
+ "stereo_corr": 1.0,
+ "darken": -0.09,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -70.5,
+ "first_sample": 0.0,
+ "tail_db": -31.0,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-815614_2.wav",
+ "length_s": 0.181,
+ "typ_z": -0.61,
+ "nn_cos": 0.841,
+ "tone_db": 10.73,
+ "tone_ratio": 1.26,
+ "n_outside": 3,
+ "outside": [
+ [
+ "31 Hz",
+ 27.0,
+ "above"
+ ],
+ [
+ "198 Hz",
+ 24.3,
+ "above"
+ ],
+ [
+ "157 Hz",
+ 20.3,
+ "above"
+ ]
+ ],
+ "attack_ms": 7.0,
+ "t40_ms": 115.0,
+ "active_s": 0.135,
+ "onset_s": 0.04,
+ "env_oct": 0.65,
+ "curve_db": 13.11,
+ "t40_ratio": 0.742,
+ "attack_ratio": 2.333,
+ "active_ratio": 0.813,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 155.0,
+ "active_s": 0.166
+ },
+ "phrase_z": 3.93,
+ "phrase_min": 3.57,
+ "margin": 2.23,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Computer keyboard",
+ 0.094
+ ],
+ [
+ "Speech",
+ 0.065
+ ],
+ [
+ "Tap",
+ 0.057
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.094,
+ "floor_db": -37.6,
+ "stereo_corr": 1.0,
+ "darken": 1.32,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -74.7,
+ "first_sample": 0.0,
+ "tail_db": -41.8,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity_bench": {
+ "n": 12,
+ "cos_mean": 0.673,
+ "cos_max": 0.919,
+ "tone_min_db": 4.64,
+ "tone_mean_db": 9.33,
+ "near_duplicates": []
+ },
+ "diversity_forge": {
+ "n": 15,
+ "cos_mean": 0.721,
+ "cos_max": 0.978,
+ "tone_min_db": 3.07,
+ "tone_mean_db": 12.39,
+ "near_duplicates": []
+ },
+ "summary": {
+ "made": {},
+ "real_cross": {
+ "n": 15,
+ "typ_z": 0.26,
+ "tone_db": 8.88,
+ "env_oct": 1.01,
+ "curve_db": 12.2,
+ "phrase_z": 4.75,
+ "margin": 2.83,
+ "as_p": 0.08,
+ "tone_ratio": 1.62,
+ "t40_ratio": 1.29,
+ "wins": 1.0,
+ "floor_db": -43.7,
+ "stereo_corr": 0.76,
+ "darken": 0.04,
+ "as_names": 0.53,
+ "artifacts": 0,
+ "var_cos": 0.721
+ },
+ "real_loo": {
+ "n": 12,
+ "typ_z": 0.23,
+ "tone_db": 5.47,
+ "env_oct": 1.52,
+ "curve_db": 16.7,
+ "phrase_z": 3.71,
+ "margin": 1.53,
+ "as_p": 0.01,
+ "tone_ratio": 1.0,
+ "t40_ratio": 1.0,
+ "wins": 1.0,
+ "floor_db": -39.65,
+ "stereo_corr": 1.0,
+ "darken": 0.09,
+ "as_names": 0.42,
+ "artifacts": 0,
+ "var_cos": 0.673
+ }
+ }
+ },
+ "whoosh_big": {
+ "n_bench": 3,
+ "n_forge": 3,
+ "note": "forge = three designed Mixkit whooshes; bench = real air / cape swoops.",
+ "bench_refs": [
+ "E:/earshot/out/refs/whoosh_big/bench/kit-whoosh_big_02.wav",
+ "E:/earshot/out/refs/whoosh_big/bench/fs-711122_1.wav",
+ "E:/earshot/out/refs/whoosh_big/bench/fs-711122_2.wav"
+ ],
+ "made": [],
+ "diversity": {
+ "n": 0
+ },
+ "real_cross": [
+ {
+ "file": "E:/earshot/out/refs/whoosh_big/forge/kit-whoosh_big_01.wav",
+ "length_s": 1.93,
+ "typ_z": 0.26,
+ "nn_cos": 0.741,
+ "tone_db": 6.27,
+ "tone_ratio": 1.48,
+ "n_outside": 10,
+ "outside": [
+ [
+ "16 kHz",
+ 13.8,
+ "above"
+ ],
+ [
+ "12.7 kHz",
+ 9.4,
+ "above"
+ ],
+ [
+ "39 Hz",
+ 7.9,
+ "above"
+ ]
+ ],
+ "attack_ms": 397.0,
+ "t40_ms": 590.0,
+ "active_s": 1.092,
+ "onset_s": 0.298,
+ "env_oct": 0.369,
+ "curve_db": 10.78,
+ "t40_ratio": 1.494,
+ "attack_ratio": 0.852,
+ "active_ratio": 1.228,
+ "ref_median": {
+ "attack_ms": 466.0,
+ "t40_ms": 395.0,
+ "active_s": 0.889
+ },
+ "phrase_z": 2.92,
+ "phrase_min": 2.64,
+ "margin": 0.04,
+ "wins": 0.5,
+ "best_confusion": "ocean waves crashing",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.573
+ ],
+ [
+ "Sound effect",
+ 0.181
+ ],
+ [
+ "Bang",
+ 0.06
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.573,
+ "floor_db": -62.1,
+ "stereo_corr": 0.879,
+ "darken": 0.3,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -141.7,
+ "first_sample": 0.0,
+ "tail_db": -104.0,
+ "lead_s": 0.049,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_big/forge/mk-1167_1.wav",
+ "length_s": 3.2,
+ "typ_z": -0.27,
+ "nn_cos": 0.488,
+ "tone_db": 8.84,
+ "tone_ratio": 2.08,
+ "n_outside": 15,
+ "outside": [
+ [
+ "16 kHz",
+ 25.0,
+ "above"
+ ],
+ [
+ "12.7 kHz",
+ 19.6,
+ "above"
+ ],
+ [
+ "10.1 kHz",
+ 14.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 908.0,
+ "t40_ms": 1890.0,
+ "active_s": 2.858,
+ "onset_s": 0.232,
+ "env_oct": 1.635,
+ "curve_db": 19.46,
+ "t40_ratio": 4.785,
+ "attack_ratio": 1.948,
+ "active_ratio": 3.215,
+ "ref_median": {
+ "attack_ms": 466.0,
+ "t40_ms": 395.0,
+ "active_s": 0.889
+ },
+ "phrase_z": 2.67,
+ "phrase_min": 2.32,
+ "margin": -1.16,
+ "wins": 0.0,
+ "best_confusion": "a strong wind storm",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.133
+ ],
+ [
+ "Sound effect",
+ 0.069
+ ],
+ [
+ "Explosion",
+ 0.054
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.133,
+ "floor_db": -32.7,
+ "stereo_corr": 0.614,
+ "darken": -0.11,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -76.9,
+ "first_sample": 0.0,
+ "tail_db": -67.5,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_big/forge/mk-1168_1.wav",
+ "length_s": 2.095,
+ "typ_z": -0.29,
+ "nn_cos": 0.497,
+ "tone_db": 8.8,
+ "tone_ratio": 2.07,
+ "n_outside": 13,
+ "outside": [
+ [
+ "6.35 kHz",
+ 19.4,
+ "above"
+ ],
+ [
+ "16 kHz",
+ 17.3,
+ "above"
+ ],
+ [
+ "8 kHz",
+ 17.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 273.0,
+ "t40_ms": 1185.0,
+ "active_s": 1.806,
+ "onset_s": 0.139,
+ "env_oct": 1.126,
+ "curve_db": 23.13,
+ "t40_ratio": 3.0,
+ "attack_ratio": 0.586,
+ "active_ratio": 2.031,
+ "ref_median": {
+ "attack_ms": 466.0,
+ "t40_ms": 395.0,
+ "active_s": 0.889
+ },
+ "phrase_z": 2.95,
+ "phrase_min": 2.57,
+ "margin": 0.81,
+ "wins": 1.0,
+ "best_confusion": "ocean waves crashing",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.521
+ ],
+ [
+ "Sound effect",
+ 0.194
+ ],
+ [
+ "Explosion",
+ 0.107
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.521,
+ "floor_db": -43.2,
+ "stereo_corr": 0.661,
+ "darken": -0.68,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -66.0,
+ "first_sample": 0.0,
+ "tail_db": -67.5,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_loo": [
+ {
+ "file": "E:/earshot/out/refs/whoosh_big/bench/kit-whoosh_big_02.wav",
+ "length_s": 1.35,
+ "nn_cos": 0.382,
+ "tone_db": 5.66,
+ "tone_ratio": 1.55,
+ "n_outside": 17,
+ "outside": [
+ [
+ "500 Hz",
+ -9.9,
+ "below"
+ ],
+ [
+ "16 kHz",
+ -9.3,
+ "below"
+ ],
+ [
+ "99 Hz",
+ 9.0,
+ "above"
+ ]
+ ],
+ "attack_ms": 466.0,
+ "t40_ms": 370.0,
+ "active_s": 0.889,
+ "onset_s": 0.211,
+ "env_oct": 0.16,
+ "curve_db": 8.0,
+ "t40_ratio": 0.886,
+ "attack_ratio": 1.223,
+ "active_ratio": 1.01,
+ "ref_median": {
+ "attack_ms": 381.0,
+ "t40_ms": 417.5,
+ "active_s": 0.88
+ },
+ "phrase_z": 2.26,
+ "phrase_min": 1.68,
+ "margin": 1.04,
+ "wins": 0.5,
+ "best_confusion": "a strong wind storm",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.698
+ ],
+ [
+ "Sound effect",
+ 0.282
+ ],
+ [
+ "Music",
+ 0.056
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.698,
+ "floor_db": -58.4,
+ "stereo_corr": 0.006,
+ "darken": -0.17,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -74.4,
+ "first_sample": 0.0,
+ "tail_db": -230.8,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_big/bench/fs-711122_1.wav",
+ "length_s": 1.23,
+ "nn_cos": 0.768,
+ "tone_db": 3.04,
+ "tone_ratio": 0.61,
+ "n_outside": 2,
+ "outside": [
+ [
+ "630 Hz",
+ 6.5,
+ "above"
+ ],
+ [
+ "8 kHz",
+ -4.8,
+ "below"
+ ]
+ ],
+ "attack_ms": 579.0,
+ "t40_ms": 395.0,
+ "active_s": 1.067,
+ "onset_s": 0.143,
+ "env_oct": 0.434,
+ "curve_db": 6.59,
+ "t40_ratio": 0.975,
+ "attack_ratio": 1.784,
+ "active_ratio": 1.349,
+ "ref_median": {
+ "attack_ms": 324.5,
+ "t40_ms": 405.0,
+ "active_s": 0.791
+ },
+ "phrase_z": 1.68,
+ "phrase_min": 0.86,
+ "margin": 0.19,
+ "wins": 0.5,
+ "best_confusion": "a strong wind storm",
+ "as_top": [
+ [
+ "Door",
+ 0.66
+ ],
+ [
+ "Slam",
+ 0.385
+ ],
+ [
+ "Sliding door",
+ 0.315
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -42.9,
+ "stereo_corr": 1.0,
+ "darken": 0.28,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -89.0,
+ "first_sample": 0.0,
+ "tail_db": -55.3,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_big/bench/fs-711122_2.wav",
+ "length_s": 0.732,
+ "nn_cos": 0.768,
+ "tone_db": 4.25,
+ "tone_ratio": 0.98,
+ "n_outside": 5,
+ "outside": [
+ [
+ "16 kHz",
+ 9.8,
+ "above"
+ ],
+ [
+ "99 Hz",
+ -7.7,
+ "below"
+ ],
+ [
+ "31 Hz",
+ -7.3,
+ "below"
+ ]
+ ],
+ "attack_ms": 183.0,
+ "t40_ms": 440.0,
+ "active_s": 0.693,
+ "onset_s": 0.002,
+ "env_oct": 0.738,
+ "curve_db": 6.67,
+ "t40_ratio": 1.15,
+ "attack_ratio": 0.35,
+ "active_ratio": 0.709,
+ "ref_median": {
+ "attack_ms": 522.5,
+ "t40_ms": 382.5,
+ "active_s": 0.978
+ },
+ "phrase_z": 1.88,
+ "phrase_min": 0.46,
+ "margin": -0.75,
+ "wins": 0.5,
+ "best_confusion": "a strong wind storm",
+ "as_top": [
+ [
+ "Door",
+ 0.176
+ ],
+ [
+ "Slam",
+ 0.143
+ ],
+ [
+ "Explosion",
+ 0.093
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.0,
+ "floor_db": -39.9,
+ "stereo_corr": 1.0,
+ "darken": 1.11,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -44.3,
+ "first_sample": 0.0,
+ "tail_db": -43.5,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_rev": [
+ {
+ "file": "E:/earshot/out/refs/whoosh_big/bench/kit-whoosh_big_02.wav",
+ "length_s": 1.35,
+ "typ_z": -8.35,
+ "nn_cos": 0.741,
+ "tone_db": 8.63,
+ "tone_ratio": 2.05,
+ "n_outside": 14,
+ "outside": [
+ [
+ "16 kHz",
+ -23.2,
+ "below"
+ ],
+ [
+ "50 Hz",
+ -16.3,
+ "below"
+ ],
+ [
+ "12.7 kHz",
+ -13.1,
+ "below"
+ ]
+ ],
+ "attack_ms": 466.0,
+ "t40_ms": 370.0,
+ "active_s": 0.889,
+ "onset_s": 0.211,
+ "env_oct": 0.978,
+ "curve_db": 22.87,
+ "t40_ratio": 0.312,
+ "attack_ratio": 1.174,
+ "active_ratio": 0.492,
+ "ref_median": {
+ "attack_ms": 397.0,
+ "t40_ms": 1185.0,
+ "active_s": 1.806
+ },
+ "phrase_z": 2.26,
+ "phrase_min": 1.68,
+ "margin": 1.04,
+ "wins": 0.5,
+ "best_confusion": "a strong wind storm",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.698
+ ],
+ [
+ "Sound effect",
+ 0.282
+ ],
+ [
+ "Music",
+ 0.056
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.698,
+ "floor_db": -58.4,
+ "stereo_corr": 0.006,
+ "darken": -0.17,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -74.4,
+ "first_sample": 0.0,
+ "tail_db": -230.8,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_big/bench/fs-711122_1.wav",
+ "length_s": 1.23,
+ "typ_z": -31.99,
+ "nn_cos": 0.399,
+ "tone_db": 8.17,
+ "tone_ratio": 1.94,
+ "n_outside": 11,
+ "outside": [
+ [
+ "16 kHz",
+ -17.3,
+ "below"
+ ],
+ [
+ "12.7 kHz",
+ -15.5,
+ "below"
+ ],
+ [
+ "8 kHz",
+ -14.8,
+ "below"
+ ]
+ ],
+ "attack_ms": 579.0,
+ "t40_ms": 395.0,
+ "active_s": 1.067,
+ "onset_s": 0.143,
+ "env_oct": 0.963,
+ "curve_db": 21.09,
+ "t40_ratio": 0.333,
+ "attack_ratio": 1.458,
+ "active_ratio": 0.591,
+ "ref_median": {
+ "attack_ms": 397.0,
+ "t40_ms": 1185.0,
+ "active_s": 1.806
+ },
+ "phrase_z": 1.68,
+ "phrase_min": 0.86,
+ "margin": 0.19,
+ "wins": 0.5,
+ "best_confusion": "a strong wind storm",
+ "as_top": [
+ [
+ "Door",
+ 0.66
+ ],
+ [
+ "Slam",
+ 0.385
+ ],
+ [
+ "Sliding door",
+ 0.315
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -42.9,
+ "stereo_corr": 1.0,
+ "darken": 0.28,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -89.0,
+ "first_sample": 0.0,
+ "tail_db": -55.3,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_big/bench/fs-711122_2.wav",
+ "length_s": 0.732,
+ "typ_z": -26.38,
+ "nn_cos": 0.45,
+ "tone_db": 6.92,
+ "tone_ratio": 1.64,
+ "n_outside": 13,
+ "outside": [
+ [
+ "50 Hz",
+ -11.7,
+ "below"
+ ],
+ [
+ "8 kHz",
+ -10.9,
+ "below"
+ ],
+ [
+ "16 kHz",
+ -10.4,
+ "below"
+ ]
+ ],
+ "attack_ms": 183.0,
+ "t40_ms": 440.0,
+ "active_s": 0.693,
+ "onset_s": 0.002,
+ "env_oct": 1.309,
+ "curve_db": 21.46,
+ "t40_ratio": 0.371,
+ "attack_ratio": 0.461,
+ "active_ratio": 0.384,
+ "ref_median": {
+ "attack_ms": 397.0,
+ "t40_ms": 1185.0,
+ "active_s": 1.806
+ },
+ "phrase_z": 1.88,
+ "phrase_min": 0.46,
+ "margin": -0.75,
+ "wins": 0.5,
+ "best_confusion": "a strong wind storm",
+ "as_top": [
+ [
+ "Door",
+ 0.176
+ ],
+ [
+ "Slam",
+ 0.143
+ ],
+ [
+ "Explosion",
+ 0.093
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.0,
+ "floor_db": -39.9,
+ "stereo_corr": 1.0,
+ "darken": 1.11,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -44.3,
+ "first_sample": 0.0,
+ "tail_db": -43.5,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity_bench": {
+ "n": 3,
+ "cos_mean": 0.495,
+ "cos_max": 0.768,
+ "tone_min_db": 3.17,
+ "tone_mean_db": 4.96,
+ "near_duplicates": []
+ },
+ "diversity_forge": {
+ "n": 3,
+ "cos_mean": 0.7,
+ "cos_max": 0.724,
+ "tone_min_db": 4.0,
+ "tone_mean_db": 5.3,
+ "near_duplicates": []
+ },
+ "summary": {
+ "made": {},
+ "real_cross": {
+ "n": 3,
+ "typ_z": -0.27,
+ "tone_db": 8.8,
+ "env_oct": 1.13,
+ "curve_db": 19.46,
+ "phrase_z": 2.92,
+ "margin": 0.04,
+ "as_p": 0.52,
+ "tone_ratio": 2.07,
+ "t40_ratio": 3.0,
+ "wins": 0.5,
+ "floor_db": -43.2,
+ "stereo_corr": 0.66,
+ "darken": -0.11,
+ "as_names": 1.0,
+ "artifacts": 0,
+ "var_cos": 0.7
+ },
+ "real_loo": {
+ "n": 3,
+ "tone_db": 4.25,
+ "env_oct": 0.43,
+ "curve_db": 6.67,
+ "phrase_z": 1.88,
+ "margin": 0.19,
+ "as_p": 0.0,
+ "tone_ratio": 0.98,
+ "t40_ratio": 0.97,
+ "wins": 0.5,
+ "floor_db": -42.9,
+ "stereo_corr": 1.0,
+ "darken": 0.28,
+ "as_names": 0.33,
+ "artifacts": 0,
+ "var_cos": 0.495
+ }
+ }
+ },
+ "whoosh_fast": {
+ "n_bench": 5,
+ "n_forge": 5,
+ "note": "",
+ "bench_refs": [
+ "E:/earshot/out/refs/whoosh_fast/bench/fs-352719_1.wav",
+ "E:/earshot/out/refs/whoosh_fast/bench/fs-216675_1.wav",
+ "E:/earshot/out/refs/whoosh_fast/bench/fs-463763_1.wav",
+ "E:/earshot/out/refs/whoosh_fast/bench/fs-463763_2.wav",
+ "E:/earshot/out/refs/whoosh_fast/bench/fs-463763_3.wav"
+ ],
+ "made": [],
+ "diversity": {
+ "n": 0
+ },
+ "real_cross": [
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/forge/kit-whoosh_fast_01.wav",
+ "length_s": 0.315,
+ "typ_z": -2.82,
+ "nn_cos": 0.805,
+ "tone_db": 10.5,
+ "tone_ratio": 2.29,
+ "n_outside": 9,
+ "outside": [
+ [
+ "10.1 kHz",
+ 14.5,
+ "above"
+ ],
+ [
+ "8 kHz",
+ 14.2,
+ "above"
+ ],
+ [
+ "12.7 kHz",
+ 14.2,
+ "above"
+ ]
+ ],
+ "attack_ms": 31.0,
+ "t40_ms": 105.0,
+ "active_s": 0.139,
+ "onset_s": 0.031,
+ "env_oct": 1.485,
+ "curve_db": 22.56,
+ "t40_ratio": 0.396,
+ "attack_ratio": 0.287,
+ "active_ratio": 0.401,
+ "ref_median": {
+ "attack_ms": 108.0,
+ "t40_ms": 265.0,
+ "active_s": 0.347
+ },
+ "phrase_z": 3.54,
+ "phrase_min": 3.04,
+ "margin": 2.29,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.27
+ ],
+ [
+ "Bang",
+ 0.108
+ ],
+ [
+ "Whack, thwack",
+ 0.09
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.007,
+ "floor_db": -65.9,
+ "stereo_corr": 0.814,
+ "darken": -0.16,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -49.0,
+ "first_sample": 0.0,
+ "tail_db": -101.0,
+ "lead_s": 0.013,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/forge/kit-whoosh_fast_02.wav",
+ "length_s": 0.5,
+ "typ_z": 1.26,
+ "nn_cos": 0.926,
+ "tone_db": 7.4,
+ "tone_ratio": 1.62,
+ "n_outside": 1,
+ "outside": [
+ [
+ "10.1 kHz",
+ 11.2,
+ "above"
+ ]
+ ],
+ "attack_ms": 72.0,
+ "t40_ms": 180.0,
+ "active_s": 0.25,
+ "onset_s": 0.09,
+ "env_oct": 0.539,
+ "curve_db": 9.22,
+ "t40_ratio": 0.679,
+ "attack_ratio": 0.667,
+ "active_ratio": 0.72,
+ "ref_median": {
+ "attack_ms": 108.0,
+ "t40_ms": 265.0,
+ "active_s": 0.347
+ },
+ "phrase_z": 4.27,
+ "phrase_min": 4.04,
+ "margin": 3.39,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Zipper (clothing)",
+ 0.343
+ ],
+ [
+ "Whoosh, swoosh, swish",
+ 0.238
+ ],
+ [
+ "Sound effect",
+ 0.118
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.238,
+ "floor_db": -64.7,
+ "stereo_corr": -0.162,
+ "darken": -0.24,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -112.1,
+ "first_sample": 0.0,
+ "tail_db": -81.3,
+ "lead_s": 0.029,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/forge/kit-whoosh_fast_03.wav",
+ "length_s": 0.39,
+ "typ_z": 1.15,
+ "nn_cos": 0.908,
+ "tone_db": 10.31,
+ "tone_ratio": 2.25,
+ "n_outside": 12,
+ "outside": [
+ [
+ "794 Hz",
+ -20.5,
+ "below"
+ ],
+ [
+ "198 Hz",
+ 18.5,
+ "above"
+ ],
+ [
+ "157 Hz",
+ 16.8,
+ "above"
+ ]
+ ],
+ "attack_ms": 51.0,
+ "t40_ms": 90.0,
+ "active_s": 0.141,
+ "onset_s": 0.094,
+ "env_oct": 1.313,
+ "curve_db": 21.2,
+ "t40_ratio": 0.34,
+ "attack_ratio": 0.472,
+ "active_ratio": 0.406,
+ "ref_median": {
+ "attack_ms": 108.0,
+ "t40_ms": 265.0,
+ "active_s": 0.347
+ },
+ "phrase_z": 4.6,
+ "phrase_min": 4.27,
+ "margin": 3.4,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Zipper (clothing)",
+ 0.211
+ ],
+ [
+ "Whoosh, swoosh, swish",
+ 0.079
+ ],
+ [
+ "Sound effect",
+ 0.048
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.079,
+ "floor_db": -65.4,
+ "stereo_corr": 0.927,
+ "darken": 0.88,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -68.5,
+ "first_sample": 0.0,
+ "tail_db": -85.1,
+ "lead_s": 0.005,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/forge/kit-whoosh_fast_04.wav",
+ "length_s": 0.395,
+ "typ_z": -2.65,
+ "nn_cos": 0.762,
+ "tone_db": 12.8,
+ "tone_ratio": 2.8,
+ "n_outside": 12,
+ "outside": [
+ [
+ "31 Hz",
+ -24.9,
+ "below"
+ ],
+ [
+ "12.7 kHz",
+ 21.4,
+ "above"
+ ],
+ [
+ "10.1 kHz",
+ 21.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 24.0,
+ "t40_ms": 160.0,
+ "active_s": 0.203,
+ "onset_s": 0.037,
+ "env_oct": 1.224,
+ "curve_db": 18.13,
+ "t40_ratio": 0.604,
+ "attack_ratio": 0.222,
+ "active_ratio": 0.585,
+ "ref_median": {
+ "attack_ms": 108.0,
+ "t40_ms": 265.0,
+ "active_s": 0.347
+ },
+ "phrase_z": 3.72,
+ "phrase_min": 3.41,
+ "margin": 2.75,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.172
+ ],
+ [
+ "Whack, thwack",
+ 0.085
+ ],
+ [
+ "Bang",
+ 0.075
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.056,
+ "floor_db": -67.3,
+ "stereo_corr": 0.919,
+ "darken": -0.54,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -76.8,
+ "first_sample": 0.0,
+ "tail_db": -86.8,
+ "lead_s": 0.004,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/forge/fs-719636_1.wav",
+ "length_s": 0.275,
+ "typ_z": 0.57,
+ "nn_cos": 0.895,
+ "tone_db": 10.03,
+ "tone_ratio": 2.19,
+ "n_outside": 2,
+ "outside": [
+ [
+ "250 Hz",
+ 16.3,
+ "above"
+ ],
+ [
+ "500 Hz",
+ -7.6,
+ "below"
+ ]
+ ],
+ "attack_ms": 22.0,
+ "t40_ms": 210.0,
+ "active_s": 0.228,
+ "onset_s": 0.017,
+ "env_oct": 1.079,
+ "curve_db": 20.2,
+ "t40_ratio": 0.792,
+ "attack_ratio": 0.204,
+ "active_ratio": 0.657,
+ "ref_median": {
+ "attack_ms": 108.0,
+ "t40_ms": 265.0,
+ "active_s": 0.347
+ },
+ "phrase_z": 4.14,
+ "phrase_min": 3.95,
+ "margin": 2.78,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.163
+ ],
+ [
+ "Bang",
+ 0.082
+ ],
+ [
+ "Whack, thwack",
+ 0.066
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.006,
+ "floor_db": -40.3,
+ "stereo_corr": 0.605,
+ "darken": 0.64,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -44.4,
+ "first_sample": 0.0,
+ "tail_db": -69.6,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_loo": [
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/bench/fs-352719_1.wav",
+ "length_s": 0.435,
+ "typ_z": -0.1,
+ "nn_cos": 0.908,
+ "tone_db": 9.54,
+ "tone_ratio": 2.14,
+ "n_outside": 5,
+ "outside": [
+ [
+ "31 Hz",
+ 23.5,
+ "above"
+ ],
+ [
+ "1 kHz",
+ -23.3,
+ "below"
+ ],
+ [
+ "39 Hz",
+ 15.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 41.0,
+ "t40_ms": 220.0,
+ "active_s": 0.325,
+ "onset_s": 0.065,
+ "env_oct": 0.601,
+ "curve_db": 8.9,
+ "t40_ratio": 0.822,
+ "attack_ratio": 0.373,
+ "active_ratio": 0.935,
+ "ref_median": {
+ "attack_ms": 110.0,
+ "t40_ms": 267.5,
+ "active_s": 0.347
+ },
+ "phrase_z": 4.44,
+ "phrase_min": 4.24,
+ "margin": 3.67,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.102
+ ],
+ [
+ "Sound effect",
+ 0.08
+ ],
+ [
+ "Zipper (clothing)",
+ 0.035
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.102,
+ "floor_db": -44.6,
+ "stereo_corr": 1.0,
+ "darken": -0.45,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -54.7,
+ "first_sample": 0.0,
+ "tail_db": -64.7,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/bench/fs-216675_1.wav",
+ "length_s": 0.23,
+ "typ_z": -0.43,
+ "nn_cos": 0.908,
+ "tone_db": 10.24,
+ "tone_ratio": 2.29,
+ "n_outside": 11,
+ "outside": [
+ [
+ "50 Hz",
+ -23.9,
+ "below"
+ ],
+ [
+ "39 Hz",
+ -17.4,
+ "below"
+ ],
+ [
+ "397 Hz",
+ 13.6,
+ "above"
+ ]
+ ],
+ "attack_ms": 33.0,
+ "t40_ms": 110.0,
+ "active_s": 0.177,
+ "onset_s": 0.023,
+ "env_oct": 1.331,
+ "curve_db": 23.4,
+ "t40_ratio": 0.411,
+ "attack_ratio": 0.3,
+ "active_ratio": 0.509,
+ "ref_median": {
+ "attack_ms": 110.0,
+ "t40_ms": 267.5,
+ "active_s": 0.347
+ },
+ "phrase_z": 4.34,
+ "phrase_min": 3.89,
+ "margin": 3.35,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Music",
+ 0.209
+ ],
+ [
+ "Sound effect",
+ 0.206
+ ],
+ [
+ "Bang",
+ 0.138
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.005,
+ "floor_db": -46.7,
+ "stereo_corr": 1.0,
+ "darken": -0.77,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -98.6,
+ "first_sample": 0.0,
+ "tail_db": -63.8,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/bench/fs-463763_1.wav",
+ "length_s": 0.435,
+ "typ_z": 1.52,
+ "nn_cos": 0.915,
+ "tone_db": 4.58,
+ "tone_ratio": 0.66,
+ "n_outside": 1,
+ "outside": [
+ [
+ "1.26 kHz",
+ 11.5,
+ "above"
+ ]
+ ],
+ "attack_ms": 117.0,
+ "t40_ms": 295.0,
+ "active_s": 0.406,
+ "onset_s": 0.009,
+ "env_oct": 0.402,
+ "curve_db": 8.21,
+ "t40_ratio": 1.216,
+ "attack_ratio": 1.57,
+ "active_ratio": 1.208,
+ "ref_median": {
+ "attack_ms": 74.5,
+ "t40_ms": 242.5,
+ "active_s": 0.336
+ },
+ "phrase_z": 3.92,
+ "phrase_min": 3.86,
+ "margin": 2.32,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.091
+ ],
+ [
+ "Whip",
+ 0.073
+ ],
+ [
+ "Sound effect",
+ 0.063
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.091,
+ "floor_db": -32.2,
+ "stereo_corr": 0.995,
+ "darken": -0.03,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -63.4,
+ "first_sample": 0.0,
+ "tail_db": -49.2,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/bench/fs-463763_2.wav",
+ "length_s": 0.398,
+ "typ_z": 6.7,
+ "nn_cos": 0.951,
+ "tone_db": 4.35,
+ "tone_ratio": 0.62,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 108.0,
+ "t40_ms": 270.0,
+ "active_s": 0.347,
+ "onset_s": 0.033,
+ "env_oct": 0.232,
+ "curve_db": 5.81,
+ "t40_ratio": 1.113,
+ "attack_ratio": 1.412,
+ "active_ratio": 1.031,
+ "ref_median": {
+ "attack_ms": 76.5,
+ "t40_ms": 242.5,
+ "active_s": 0.337
+ },
+ "phrase_z": 4.14,
+ "phrase_min": 4.01,
+ "margin": 2.62,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.138
+ ],
+ [
+ "Sound effect",
+ 0.063
+ ],
+ [
+ "Music",
+ 0.033
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.138,
+ "floor_db": -28.7,
+ "stereo_corr": 0.991,
+ "darken": -0.18,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -51.9,
+ "first_sample": 0.0,
+ "tail_db": -49.3,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/bench/fs-463763_3.wav",
+ "length_s": 0.392,
+ "typ_z": 3.93,
+ "nn_cos": 0.951,
+ "tone_db": 4.26,
+ "tone_ratio": 0.6,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 112.0,
+ "t40_ms": 265.0,
+ "active_s": 0.348,
+ "onset_s": 0.032,
+ "env_oct": 0.251,
+ "curve_db": 4.29,
+ "t40_ratio": 1.082,
+ "attack_ratio": 1.503,
+ "active_ratio": 1.036,
+ "ref_median": {
+ "attack_ms": 74.5,
+ "t40_ms": 245.0,
+ "active_s": 0.336
+ },
+ "phrase_z": 4.03,
+ "phrase_min": 3.94,
+ "margin": 2.09,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.093
+ ],
+ [
+ "Sound effect",
+ 0.04
+ ],
+ [
+ "Music",
+ 0.024
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.093,
+ "floor_db": -28.9,
+ "stereo_corr": 0.996,
+ "darken": -0.2,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -67.1,
+ "first_sample": 0.0,
+ "tail_db": -47.8,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_rev": [
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/bench/fs-352719_1.wav",
+ "length_s": 0.435,
+ "typ_z": 1.44,
+ "nn_cos": 0.897,
+ "tone_db": 7.49,
+ "tone_ratio": 0.99,
+ "n_outside": 2,
+ "outside": [
+ [
+ "157 Hz",
+ -8.1,
+ "below"
+ ],
+ [
+ "2.52 kHz",
+ -6.5,
+ "below"
+ ]
+ ],
+ "attack_ms": 41.0,
+ "t40_ms": 220.0,
+ "active_s": 0.325,
+ "onset_s": 0.065,
+ "env_oct": 0.514,
+ "curve_db": 15.4,
+ "t40_ratio": 1.375,
+ "attack_ratio": 1.323,
+ "active_ratio": 1.601,
+ "ref_median": {
+ "attack_ms": 31.0,
+ "t40_ms": 160.0,
+ "active_s": 0.203
+ },
+ "phrase_z": 4.44,
+ "phrase_min": 4.24,
+ "margin": 3.67,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.102
+ ],
+ [
+ "Sound effect",
+ 0.08
+ ],
+ [
+ "Zipper (clothing)",
+ 0.035
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.102,
+ "floor_db": -44.6,
+ "stereo_corr": 1.0,
+ "darken": -0.45,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -54.7,
+ "first_sample": 0.0,
+ "tail_db": -64.7,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/bench/fs-216675_1.wav",
+ "length_s": 0.23,
+ "typ_z": 2.39,
+ "nn_cos": 0.926,
+ "tone_db": 8.07,
+ "tone_ratio": 1.07,
+ "n_outside": 4,
+ "outside": [
+ [
+ "50 Hz",
+ -29.1,
+ "below"
+ ],
+ [
+ "39 Hz",
+ -14.3,
+ "below"
+ ],
+ [
+ "2.52 kHz",
+ -5.3,
+ "below"
+ ]
+ ],
+ "attack_ms": 33.0,
+ "t40_ms": 110.0,
+ "active_s": 0.177,
+ "onset_s": 0.023,
+ "env_oct": 0.276,
+ "curve_db": 7.54,
+ "t40_ratio": 0.688,
+ "attack_ratio": 1.065,
+ "active_ratio": 0.872,
+ "ref_median": {
+ "attack_ms": 31.0,
+ "t40_ms": 160.0,
+ "active_s": 0.203
+ },
+ "phrase_z": 4.34,
+ "phrase_min": 3.89,
+ "margin": 3.35,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Music",
+ 0.209
+ ],
+ [
+ "Sound effect",
+ 0.206
+ ],
+ [
+ "Bang",
+ 0.138
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.005,
+ "floor_db": -46.7,
+ "stereo_corr": 1.0,
+ "darken": -0.77,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -98.6,
+ "first_sample": 0.0,
+ "tail_db": -63.8,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/bench/fs-463763_1.wav",
+ "length_s": 0.435,
+ "typ_z": -6.16,
+ "nn_cos": 0.698,
+ "tone_db": 8.45,
+ "tone_ratio": 1.12,
+ "n_outside": 10,
+ "outside": [
+ [
+ "1 kHz",
+ 20.5,
+ "above"
+ ],
+ [
+ "1.26 kHz",
+ 14.9,
+ "above"
+ ],
+ [
+ "794 Hz",
+ 14.7,
+ "above"
+ ]
+ ],
+ "attack_ms": 117.0,
+ "t40_ms": 295.0,
+ "active_s": 0.406,
+ "onset_s": 0.009,
+ "env_oct": 1.266,
+ "curve_db": 23.08,
+ "t40_ratio": 1.844,
+ "attack_ratio": 3.774,
+ "active_ratio": 2.0,
+ "ref_median": {
+ "attack_ms": 31.0,
+ "t40_ms": 160.0,
+ "active_s": 0.203
+ },
+ "phrase_z": 3.92,
+ "phrase_min": 3.86,
+ "margin": 2.32,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.091
+ ],
+ [
+ "Whip",
+ 0.073
+ ],
+ [
+ "Sound effect",
+ 0.063
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.091,
+ "floor_db": -32.2,
+ "stereo_corr": 0.995,
+ "darken": -0.03,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -63.4,
+ "first_sample": 0.0,
+ "tail_db": -49.2,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/bench/fs-463763_2.wav",
+ "length_s": 0.398,
+ "typ_z": -3.76,
+ "nn_cos": 0.745,
+ "tone_db": 8.15,
+ "tone_ratio": 1.08,
+ "n_outside": 9,
+ "outside": [
+ [
+ "1 kHz",
+ 20.6,
+ "above"
+ ],
+ [
+ "794 Hz",
+ 14.8,
+ "above"
+ ],
+ [
+ "10.1 kHz",
+ -13.7,
+ "below"
+ ]
+ ],
+ "attack_ms": 108.0,
+ "t40_ms": 270.0,
+ "active_s": 0.347,
+ "onset_s": 0.033,
+ "env_oct": 1.11,
+ "curve_db": 22.91,
+ "t40_ratio": 1.688,
+ "attack_ratio": 3.484,
+ "active_ratio": 1.709,
+ "ref_median": {
+ "attack_ms": 31.0,
+ "t40_ms": 160.0,
+ "active_s": 0.203
+ },
+ "phrase_z": 4.14,
+ "phrase_min": 4.01,
+ "margin": 2.62,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.138
+ ],
+ [
+ "Sound effect",
+ 0.063
+ ],
+ [
+ "Music",
+ 0.033
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.138,
+ "floor_db": -28.7,
+ "stereo_corr": 0.991,
+ "darken": -0.18,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -51.9,
+ "first_sample": 0.0,
+ "tail_db": -49.3,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/bench/fs-463763_3.wav",
+ "length_s": 0.392,
+ "typ_z": -4.94,
+ "nn_cos": 0.721,
+ "tone_db": 8.21,
+ "tone_ratio": 1.09,
+ "n_outside": 9,
+ "outside": [
+ [
+ "1 kHz",
+ 21.5,
+ "above"
+ ],
+ [
+ "794 Hz",
+ 15.0,
+ "above"
+ ],
+ [
+ "10.1 kHz",
+ -12.9,
+ "below"
+ ]
+ ],
+ "attack_ms": 112.0,
+ "t40_ms": 265.0,
+ "active_s": 0.348,
+ "onset_s": 0.032,
+ "env_oct": 1.12,
+ "curve_db": 21.43,
+ "t40_ratio": 1.656,
+ "attack_ratio": 3.613,
+ "active_ratio": 1.714,
+ "ref_median": {
+ "attack_ms": 31.0,
+ "t40_ms": 160.0,
+ "active_s": 0.203
+ },
+ "phrase_z": 4.03,
+ "phrase_min": 3.94,
+ "margin": 2.09,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.093
+ ],
+ [
+ "Sound effect",
+ 0.04
+ ],
+ [
+ "Music",
+ 0.024
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.093,
+ "floor_db": -28.9,
+ "stereo_corr": 0.996,
+ "darken": -0.2,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -67.1,
+ "first_sample": 0.0,
+ "tail_db": -47.8,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity_bench": {
+ "n": 5,
+ "cos_mean": 0.772,
+ "cos_max": 0.951,
+ "tone_min_db": 2.28,
+ "tone_mean_db": 8.01,
+ "near_duplicates": []
+ },
+ "diversity_forge": {
+ "n": 5,
+ "cos_mean": 0.826,
+ "cos_max": 0.892,
+ "tone_min_db": 5.21,
+ "tone_mean_db": 10.47,
+ "near_duplicates": []
+ },
+ "summary": {
+ "made": {},
+ "real_cross": {
+ "n": 5,
+ "typ_z": 0.57,
+ "tone_db": 10.31,
+ "env_oct": 1.22,
+ "curve_db": 20.2,
+ "phrase_z": 4.14,
+ "margin": 2.78,
+ "as_p": 0.06,
+ "tone_ratio": 2.25,
+ "t40_ratio": 0.6,
+ "wins": 1.0,
+ "floor_db": -65.4,
+ "stereo_corr": 0.81,
+ "darken": -0.16,
+ "as_names": 0.4,
+ "artifacts": 0,
+ "var_cos": 0.826
+ },
+ "real_loo": {
+ "n": 5,
+ "typ_z": 1.52,
+ "tone_db": 4.58,
+ "env_oct": 0.4,
+ "curve_db": 8.21,
+ "phrase_z": 4.14,
+ "margin": 2.62,
+ "as_p": 0.09,
+ "tone_ratio": 0.66,
+ "t40_ratio": 1.08,
+ "wins": 1.0,
+ "floor_db": -32.2,
+ "stereo_corr": 1.0,
+ "darken": -0.2,
+ "as_names": 0.8,
+ "artifacts": 0,
+ "var_cos": 0.772
+ }
+ }
+ }
+ },
+ "seconds": 19.4
+}
\ No newline at end of file
diff --git a/eval/forge/results_real.md b/eval/forge/results_real.md
new file mode 100644
index 0000000..9cb2bd1
--- /dev/null
+++ b/eval/forge/results_real.md
@@ -0,0 +1,87 @@
+# forge benchmark: real
+
+made: -; refs: E:\earshot\.cache\refsets.json; 2026-10-16 04:07
+
+Medians per set. made = the made sounds vs the bench refs; real x = the forge refs (real, never in the bench) vs the bench refs; real loo = each bench ref vs the other bench refs. typ z, phrase z, margin, AS names: higher is better; tone dB, env oct (octaves off in attack / t40 / length), curve dB: lower is better; t40 x = tail length / bench median (1 = same). inside = checks (of n) a made sound lands inside the real 80% limit, median over the made sounds (real = the same count for the real rows). var cos = mean pairwise CLAP cosine within the set (made variants / forge refs / bench refs): high = the variants are near clones.
+
+| category | set | n | typ z | tone dB | env oct | curve dB | t40 x | phrase z | margin | AS names | artifacts | var cos | inside |
+|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
+| click | real x | 8 | 0.16 | 11.33 | 0.22 | 9.92 | 1.33 | 2.77 | 0.53 | 0.12 | 0 | 0.651 | |
+| | real loo | 7 | -0.1 | 9.85 | 0.46 | 13.19 | 0.86 | 2.67 | 0.27 | 0 | 0 | 0.576 | |
+| downlifter | no rows | 0 | | | | | | | | | | | |
+| | note: empty: the licensed library has no downlifter, real or designed (no reversing or pitching of other sounds here: trim + fade only). Benchmark made downlifters on phrasings, AudioSet and artifacts only. | | | | | | | | | | | | |
+| glitch | no rows | 0 | | | | | | | | | | | |
+| | note: empty: no glitch material, real or designed, in the licensed library. Benchmark made glitches on phrasings, AudioSet and artifacts only. | | | | | | | | | | | | |
+| hover_tick | real x | 6 | 0.09 | 21.2 | 1.64 | 9.88 | 0.9 | 2.51 | 0.9 | 0 | 0 | 0.799 | |
+| | real loo | 6 | 0.47 | 15.46 | 1.16 | 19.39 | 1.2 | 2 | 0.58 | 0 | 4 | 0.626 | |
+| impact | real x | 2 | | 6.19 | 2.5 | 31.98 | 0.46 | 3.04 | 1.09 | 0.5 | 0 | 0.659 | |
+| | note: thin: the library has three cinematic hits (all three were the v1 impact forge refs); one is held out as bench, so v1 impact bench numbers are optimistic. | | | | | | | | | | | | |
+| logo_sting | real x | 2 | | 15.52 | 1.46 | 21.27 | 1.7 | 1.4 | -0.82 | 0.5 | 0 | 0.185 | |
+| | real loo | 2 | | 23.22 | 0.81 | 8.99 | 1.12 | 1.36 | -0.58 | 1 | 0 | 0.507 | |
+| | note: no real logo stings exist in the library (they are composed); the refs are the struck metal / glass rings a sting ends on: a tail-shape and tone reference only. | | | | | | | | | | | | |
+| notify | real x | 5 | -0.24 | 21.3 | 1.3 | 28.92 | 0.34 | 1.25 | -0.24 | 1 | 0 | 0.593 | |
+| | real loo | 5 | 3.84 | 16.7 | 0.81 | 17.41 | 1.19 | 1.76 | -0.91 | 1 | 0 | 0.476 | |
+| | note: struck bells / glass / kalimba (forge) and celesta / glass / triangle (bench), rings capped at 2 s with a 0.25 s fade; no designed UI dings in the library. | | | | | | | | | | | | |
+| pop | real x | 7 | -0.09 | 16.06 | 1.51 | 20.54 | 0.91 | 2.46 | -0.5 | 0.29 | 0 | 0.501 | |
+| | real loo | 6 | 1.13 | 13.66 | 0.41 | 12.5 | 1.02 | 1.96 | -1.06 | 0.67 | 0 | 0.522 | |
+| riser | real x | 1 | | 17.61 | 1.01 | 20.04 | 0.5 | 3.52 | 1.06 | 1 | 0 | | |
+| | real loo | 2 | | 13.22 | 0.95 | 36.07 | 3.35 | 2.7 | 1.15 | 0.5 | 0 | 0.767 | |
+| | note: thin: one designed trailer riser (forge); the bench is two real cymbal swells, no synth risers. | | | | | | | | | | | | |
+| shimmer | real x | 5 | 1.69 | 9.66 | 0.73 | 8.58 | 0.73 | 2.87 | 0.03 | 0.6 | 2 | 0.59 | |
+| | real loo | 4 | 1.01 | 15.52 | 1.33 | 12.89 | 1.05 | 3.4 | 0.56 | 1 | 0 | 0.644 | |
+| | note: kit shimmer_01 (bar chimes sweep) is bench: its author also recorded the celesta notify bench. | | | | | | | | | | | | |
+| sub_drop | no rows | 0 | | | | | | | | | | | |
+| | note: one designed source (a tom pitched down 24 semitones by its author), held out as bench; the forge has no sub-drop refs. | | | | | | | | | | | | |
+| swipe | real x | 4 | -10.07 | 16.83 | 0.43 | 14.12 | 0.79 | 2.7 | 0.59 | 1 | 0 | 0.845 | |
+| | real loo | 4 | 1.25 | 3.64 | 0.22 | 10.07 | 1.02 | 3.42 | 2.08 | 0.25 | 0 | 0.857 | |
+| | note: cloth only: the library has no paper slides. | | | | | | | | | | | | |
+| swish | real x | 7 | 1.94 | 11.08 | 0.64 | 12.88 | 0.64 | 3.22 | 1.6 | 0.57 | 0 | 0.811 | |
+| | real loo | 3 | | 9.01 | 0.91 | 16.45 | 1.16 | 3.5 | 1.92 | 0.67 | 0 | 0.79 | |
+| tap | real x | 7 | -1.17 | 9.35 | 1.59 | 13.67 | 1.64 | 1.85 | 0.92 | 0.14 | 0 | 0.713 | |
+| | real loo | 6 | 1.66 | 16.78 | 1.19 | 11.59 | 1 | 1.62 | 0.35 | 0.17 | 0 | 0.716 | |
+| toggle | real x | 4 | -2.84 | 12.48 | 1.87 | 18.78 | 2.11 | 2.1 | -0.02 | 0 | 0 | 0.477 | |
+| | real loo | 4 | 1.54 | 6.25 | 0.28 | 4.92 | 1.02 | 1.94 | -0.28 | 0 | 0 | 0.725 | |
+| typing_key | real x | 15 | 0.26 | 8.88 | 1.01 | 12.2 | 1.29 | 4.75 | 2.83 | 0.53 | 0 | 0.721 | |
+| | real loo | 12 | 0.23 | 5.47 | 1.52 | 16.7 | 1 | 3.71 | 1.53 | 0.42 | 0 | 0.673 | |
+| whoosh_big | real x | 3 | -0.27 | 8.8 | 1.13 | 19.46 | 3 | 2.92 | 0.04 | 1 | 0 | 0.7 | |
+| | real loo | 3 | | 4.25 | 0.43 | 6.67 | 0.97 | 1.88 | 0.19 | 0.33 | 0 | 0.495 | |
+| | note: forge = three designed Mixkit whooshes; bench = real air / cape swoops. | | | | | | | | | | | | |
+| whoosh_fast | real x | 5 | 0.57 | 10.31 | 1.22 | 20.2 | 0.6 | 4.14 | 2.78 | 0.4 | 0 | 0.826 | |
+| | real loo | 5 | 1.52 | 4.58 | 0.4 | 8.21 | 1.08 | 4.14 | 2.62 | 0.8 | 0 | 0.772 | |
+
+Texture (medians; model-free, not in any forge objective): floor dB = 10th percentile 5 ms frame re the loudest (-120 = digital silence), stereo corr (1 = mono), darken = log2 centroid at half the tail / at the peak (< 0: the decay darkens).
+
+| category | set | floor dB | stereo corr | darken |
+|---|---|---|---|---|
+| | real x | -42.85 | 0.99 | -0.12 |
+| | real loo | -44.8 | 1.0 | -0.07 |
+| | real x | -43.55 | 1.0 | -0.09 |
+| | real loo | -24.9 | 0.52 | -0.01 |
+| | real x | -70.55 | 0.54 | -1.3 |
+| | real x | -52.1 | 0.59 | -0.85 |
+| | real loo | -44.1 | 0.81 | -1.43 |
+| | real x | -41.2 | 0.99 | -1.06 |
+| | real loo | -24.4 | 0.61 | -0.53 |
+| | real x | -53.2 | 1.0 | 0.21 |
+| | real loo | -47.2 | 1.0 | 0.12 |
+| | real x | -49.9 | 0.0 | |
+| | real loo | -30.65 | 1.0 | 0.12 |
+| | real x | -46.8 | 0.17 | -0.29 |
+| | real loo | -36.1 | 0.68 | -0.33 |
+| | real x | -56.85 | 0.42 | 0.07 |
+| | real loo | -38.05 | 0.68 | 0.61 |
+| | real x | -55.6 | 0.87 | -0.16 |
+| | real loo | -44.6 | 1.0 | -0.45 |
+| | real x | -43.6 | 1.0 | 1.31 |
+| | real loo | -38.4 | 1.0 | 0.94 |
+| | real x | -48.4 | 1.0 | -0.11 |
+| | real loo | -49.1 | 0.48 | -0.18 |
+| | real x | -43.7 | 0.76 | 0.04 |
+| | real loo | -39.65 | 1.0 | 0.09 |
+| | real x | -43.2 | 0.66 | -0.11 |
+| | real loo | -42.9 | 1.0 | 0.28 |
+| | real x | -65.4 | 0.81 | -0.16 |
+| | real loo | -32.2 | 1.0 | -0.2 |
+
+Diversity among the made variants (CLAP cosine mean / max, tonal distance min dB, near duplicates):
+
diff --git a/eval/forge/results_v1.json b/eval/forge/results_v1.json
new file mode 100644
index 0000000..86964f3
--- /dev/null
+++ b/eval/forge/results_v1.json
@@ -0,0 +1,8475 @@
+{
+ "tag": "v1",
+ "made_from": [
+ "typing_key=demo/forge-test/keypress/final/*.wav",
+ "click=demo/forge-test/click/final/*.wav",
+ "pop=demo/forge-test/pop/final/*.wav",
+ "whoosh_fast=demo/forge-test/whoosh/final/*.wav",
+ "impact=demo/forge-test/impact/final/*.wav"
+ ],
+ "registry": "E:\\earshot\\.cache\\refsets.json",
+ "created": "2026-10-16 04:11",
+ "phrases": {
+ "click": [
+ "a crisp tiny plastic click",
+ "a short clean click of a button being pressed"
+ ],
+ "impact": [
+ "a huge low thud for a movie trailer",
+ "a massive hit with a rumbling tail"
+ ],
+ "pop": [
+ "a popping sound like a bubble bursting",
+ "a quick plop"
+ ],
+ "typing_key": [
+ "one key being pressed on a computer keyboard",
+ "a single keystroke on a mechanical keyboard"
+ ],
+ "whoosh_fast": [
+ "a quick swoosh as something flies past",
+ "a short sharp swoosh of air"
+ ]
+ },
+ "confusions": {
+ "click": [
+ "a wooden stick hit",
+ "a finger snap",
+ "a drop of water",
+ "a stapler"
+ ],
+ "impact": [
+ "a gunshot",
+ "a door being slammed",
+ "a bass drum",
+ "distant fireworks"
+ ],
+ "pop": [
+ "a hand clap",
+ "a gunshot",
+ "a kiss",
+ "a tongue click"
+ ],
+ "typing_key": [
+ "a ballpoint pen click",
+ "a stapler",
+ "a light switch",
+ "a dripping tap"
+ ],
+ "whoosh_fast": [
+ "a steam hiss",
+ "a car driving past",
+ "tearing paper",
+ "a bird flapping its wings"
+ ]
+ },
+ "as_labels": {
+ "click": [
+ "Clicking",
+ "Tick",
+ "Tap"
+ ],
+ "impact": [
+ "Boom",
+ "Thump, thud",
+ "Explosion",
+ "Bang",
+ "Bass drum"
+ ],
+ "pop": [
+ "Burst, pop",
+ "Plop"
+ ],
+ "typing_key": [
+ "Computer keyboard",
+ "Typing",
+ "Typewriter",
+ "Clicking"
+ ],
+ "whoosh_fast": [
+ "Whoosh, swoosh, swish",
+ "Whip"
+ ]
+ },
+ "categories": {
+ "click": {
+ "n_bench": 7,
+ "n_forge": 8,
+ "note": "",
+ "bench_refs": [
+ "E:/earshot/out/refs/click/bench/mk-1124_1.wav",
+ "E:/earshot/out/refs/click/bench/mk-2364_1.wav",
+ "E:/earshot/out/refs/click/bench/fs-610524_1.wav",
+ "E:/earshot/out/refs/click/bench/fs-461799_1.wav",
+ "E:/earshot/out/refs/click/bench/fs-461799_2.wav",
+ "E:/earshot/out/refs/click/bench/fs-431419_1.wav",
+ "E:/earshot/out/refs/click/bench/fs-431419_2.wav"
+ ],
+ "made": [
+ {
+ "file": "demo/forge-test/click/final/click_01.wav",
+ "length_s": 0.078,
+ "typ_z": 2.75,
+ "nn_cos": 0.818,
+ "tone_db": 7.63,
+ "tone_ratio": 0.77,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 1.0,
+ "t40_ms": 25.0,
+ "active_s": 0.036,
+ "onset_s": 0.004,
+ "env_oct": 0.609,
+ "curve_db": 9.37,
+ "t40_ratio": 0.556,
+ "attack_ratio": 1.0,
+ "active_ratio": 0.507,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.071
+ },
+ "phrase_z": 3.39,
+ "phrase_min": 3.3,
+ "margin": 1.39,
+ "wins": 1.0,
+ "best_confusion": "a finger snap",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.053
+ ],
+ [
+ "Speech",
+ 0.036
+ ],
+ [
+ "Music",
+ 0.035
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.003,
+ "floor_db": -49.0,
+ "stereo_corr": 1.0,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -67.0,
+ "first_sample": 0.0,
+ "tail_db": -57.6,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "demo/forge-test/click/final/click_02.wav",
+ "length_s": 0.094,
+ "typ_z": 1.39,
+ "nn_cos": 0.757,
+ "tone_db": 7.6,
+ "tone_ratio": 0.77,
+ "n_outside": 1,
+ "outside": [
+ [
+ "1.26 kHz",
+ 11.3,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 40.0,
+ "active_s": 0.041,
+ "onset_s": 0.004,
+ "env_oct": 0.321,
+ "curve_db": 8.04,
+ "t40_ratio": 0.889,
+ "attack_ratio": 1.0,
+ "active_ratio": 0.577,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.071
+ },
+ "phrase_z": 3.12,
+ "phrase_min": 2.97,
+ "margin": 1.13,
+ "wins": 1.0,
+ "best_confusion": "a finger snap",
+ "as_top": [
+ [
+ "Speech",
+ 0.123
+ ],
+ [
+ "Music",
+ 0.064
+ ],
+ [
+ "Sound effect",
+ 0.023
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.004,
+ "floor_db": -46.3,
+ "stereo_corr": 1.0,
+ "darken": 0.41,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -79.9,
+ "first_sample": 0.0,
+ "tail_db": -58.0,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "demo/forge-test/click/final/click_03.wav",
+ "length_s": 0.082,
+ "typ_z": 1.89,
+ "nn_cos": 0.8,
+ "tone_db": 5.4,
+ "tone_ratio": 0.55,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 1.0,
+ "t40_ms": 25.0,
+ "active_s": 0.04,
+ "onset_s": 0.005,
+ "env_oct": 0.559,
+ "curve_db": 8.85,
+ "t40_ratio": 0.556,
+ "attack_ratio": 1.0,
+ "active_ratio": 0.563,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.071
+ },
+ "phrase_z": 3.07,
+ "phrase_min": 2.88,
+ "margin": 0.99,
+ "wins": 1.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.057
+ ],
+ [
+ "Speech",
+ 0.04
+ ],
+ [
+ "Music",
+ 0.037
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -52.2,
+ "stereo_corr": 1.0,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -87.5,
+ "first_sample": 0.0,
+ "tail_db": -57.7,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity": {
+ "n": 3,
+ "cos_mean": 0.886,
+ "cos_max": 0.93,
+ "tone_min_db": 7.24,
+ "tone_mean_db": 8.13,
+ "near_duplicates": []
+ },
+ "real_cross": [
+ {
+ "file": "E:/earshot/out/refs/click/forge/kit-ui_click_01.wav",
+ "length_s": 0.096,
+ "typ_z": -2.17,
+ "nn_cos": 0.621,
+ "tone_db": 20.26,
+ "tone_ratio": 2.06,
+ "n_outside": 13,
+ "outside": [
+ [
+ "31 Hz",
+ 44.1,
+ "above"
+ ],
+ [
+ "4 kHz",
+ -32.4,
+ "below"
+ ],
+ [
+ "3.17 kHz",
+ -30.5,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 60.0,
+ "active_s": 0.082,
+ "onset_s": 0.008,
+ "env_oct": 0.208,
+ "curve_db": 9.91,
+ "t40_ratio": 1.333,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.155,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.071
+ },
+ "phrase_z": 2.54,
+ "phrase_min": 2.14,
+ "margin": 1.05,
+ "wins": 0.5,
+ "best_confusion": "a drop of water",
+ "as_top": [
+ [
+ "Speech",
+ 0.125
+ ],
+ [
+ "Vehicle",
+ 0.101
+ ],
+ [
+ "Music",
+ 0.094
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -43.6,
+ "stereo_corr": 1.0,
+ "darken": 1.51,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -52.9,
+ "first_sample": 0.0,
+ "tail_db": -39.4,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/forge/kit-ui_click_02.wav",
+ "length_s": 0.118,
+ "typ_z": 0.04,
+ "nn_cos": 0.72,
+ "tone_db": 7.78,
+ "tone_ratio": 0.79,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 1.0,
+ "t40_ms": 110.0,
+ "active_s": 0.107,
+ "onset_s": 0.008,
+ "env_oct": 0.627,
+ "curve_db": 14.28,
+ "t40_ratio": 2.444,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.507,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.071
+ },
+ "phrase_z": 2.37,
+ "phrase_min": 2.33,
+ "margin": 0.07,
+ "wins": 0.0,
+ "best_confusion": "a finger snap",
+ "as_top": [
+ [
+ "Speech",
+ 0.119
+ ],
+ [
+ "Slap, smack",
+ 0.047
+ ],
+ [
+ "Music",
+ 0.039
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.012,
+ "floor_db": -35.1,
+ "stereo_corr": 0.916,
+ "darken": 0.23,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -64.3,
+ "first_sample": 0.0,
+ "tail_db": -38.5,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/forge/kit-ui_click_03.wav",
+ "length_s": 0.159,
+ "typ_z": 0.94,
+ "nn_cos": 0.755,
+ "tone_db": 21.05,
+ "tone_ratio": 2.14,
+ "n_outside": 15,
+ "outside": [
+ [
+ "630 Hz",
+ -35.9,
+ "below"
+ ],
+ [
+ "500 Hz",
+ -33.4,
+ "below"
+ ],
+ [
+ "99 Hz",
+ -33.3,
+ "below"
+ ]
+ ],
+ "attack_ms": 5.0,
+ "t40_ms": 30.0,
+ "active_s": 0.032,
+ "onset_s": 0.008,
+ "env_oct": 1.352,
+ "curve_db": 10.15,
+ "t40_ratio": 0.667,
+ "attack_ratio": 5.0,
+ "active_ratio": 0.451,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.071
+ },
+ "phrase_z": 2.97,
+ "phrase_min": 2.9,
+ "margin": 0.25,
+ "wins": 1.0,
+ "best_confusion": "a finger snap",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.107
+ ],
+ [
+ "Slap, smack",
+ 0.095
+ ],
+ [
+ "Music",
+ 0.045
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.004,
+ "floor_db": -58.0,
+ "stereo_corr": 0.973,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -112.3,
+ "first_sample": 0.0,
+ "tail_db": -59.4,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/forge/kit-ui_click_04.wav",
+ "length_s": 0.072,
+ "typ_z": 0.29,
+ "nn_cos": 0.702,
+ "tone_db": 12.86,
+ "tone_ratio": 1.31,
+ "n_outside": 7,
+ "outside": [
+ [
+ "16 kHz",
+ -37.5,
+ "below"
+ ],
+ [
+ "12.7 kHz",
+ -25.9,
+ "below"
+ ],
+ [
+ "39 Hz",
+ 20.5,
+ "above"
+ ]
+ ],
+ "attack_ms": 3.0,
+ "t40_ms": 55.0,
+ "active_s": 0.058,
+ "onset_s": 0.007,
+ "env_oct": 0.722,
+ "curve_db": 9.93,
+ "t40_ratio": 1.222,
+ "attack_ratio": 3.0,
+ "active_ratio": 0.817,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.071
+ },
+ "phrase_z": 2.77,
+ "phrase_min": 2.41,
+ "margin": 0.52,
+ "wins": 0.5,
+ "best_confusion": "a wooden stick hit",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.069
+ ],
+ [
+ "Music",
+ 0.024
+ ],
+ [
+ "Speech",
+ 0.024
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -34.0,
+ "stereo_corr": 1.0,
+ "darken": -0.63,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -54.2,
+ "first_sample": 0.0,
+ "tail_db": -37.4,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/forge/fs-167829_1.wav",
+ "length_s": 0.072,
+ "typ_z": 1.72,
+ "nn_cos": 0.807,
+ "tone_db": 6.17,
+ "tone_ratio": 0.63,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 1.0,
+ "t40_ms": 55.0,
+ "active_s": 0.055,
+ "onset_s": 0.005,
+ "env_oct": 0.219,
+ "curve_db": 8.44,
+ "t40_ratio": 1.222,
+ "attack_ratio": 1.0,
+ "active_ratio": 0.775,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.071
+ },
+ "phrase_z": 2.7,
+ "phrase_min": 2.55,
+ "margin": -0.27,
+ "wins": 0.0,
+ "best_confusion": "a finger snap",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.062
+ ],
+ [
+ "Speech",
+ 0.028
+ ],
+ [
+ "Music",
+ 0.028
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -41.5,
+ "stereo_corr": 0.919,
+ "darken": 0.02,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -69.7,
+ "first_sample": 0.0,
+ "tail_db": -48.4,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/forge/fs-167829_2.wav",
+ "length_s": 0.074,
+ "typ_z": 1.5,
+ "nn_cos": 0.769,
+ "tone_db": 7.53,
+ "tone_ratio": 0.76,
+ "n_outside": 3,
+ "outside": [
+ [
+ "62 Hz",
+ 17.4,
+ "above"
+ ],
+ [
+ "50 Hz",
+ 13.9,
+ "above"
+ ],
+ [
+ "39 Hz",
+ 13.5,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 60.0,
+ "active_s": 0.06,
+ "onset_s": 0.005,
+ "env_oct": 0.219,
+ "curve_db": 8.61,
+ "t40_ratio": 1.333,
+ "attack_ratio": 1.0,
+ "active_ratio": 0.845,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.071
+ },
+ "phrase_z": 2.77,
+ "phrase_min": 2.65,
+ "margin": 0.92,
+ "wins": 1.0,
+ "best_confusion": "a wooden stick hit",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.068
+ ],
+ [
+ "Speech",
+ 0.03
+ ],
+ [
+ "Music",
+ 0.024
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -42.1,
+ "stereo_corr": 0.934,
+ "darken": -0.12,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -82.5,
+ "first_sample": 0.0,
+ "tail_db": -45.8,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/forge/fs-383863_1.wav",
+ "length_s": 0.165,
+ "typ_z": -0.36,
+ "nn_cos": 0.673,
+ "tone_db": 11.27,
+ "tone_ratio": 1.14,
+ "n_outside": 3,
+ "outside": [
+ [
+ "16 kHz",
+ -37.5,
+ "below"
+ ],
+ [
+ "315 Hz",
+ 15.0,
+ "above"
+ ],
+ [
+ "1.26 kHz",
+ 12.7,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 70.0,
+ "active_s": 0.072,
+ "onset_s": 0.073,
+ "env_oct": 0.219,
+ "curve_db": 7.33,
+ "t40_ratio": 1.556,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.014,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.071
+ },
+ "phrase_z": 2.88,
+ "phrase_min": 2.62,
+ "margin": 1.11,
+ "wins": 1.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.113
+ ],
+ [
+ "Sound effect",
+ 0.052
+ ],
+ [
+ "Tap",
+ 0.037
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.037,
+ "floor_db": -56.4,
+ "stereo_corr": 1.0,
+ "darken": -1.02,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -85.2,
+ "first_sample": 0.0,
+ "tail_db": -47.4,
+ "lead_s": 0.057,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/forge/fs-383863_2.wav",
+ "length_s": 0.18,
+ "typ_z": -1.1,
+ "nn_cos": 0.649,
+ "tone_db": 11.39,
+ "tone_ratio": 1.16,
+ "n_outside": 4,
+ "outside": [
+ [
+ "16 kHz",
+ -37.5,
+ "below"
+ ],
+ [
+ "2.52 kHz",
+ -15.1,
+ "below"
+ ],
+ [
+ "3.17 kHz",
+ -11.8,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 70.0,
+ "active_s": 0.071,
+ "onset_s": 0.089,
+ "env_oct": 0.212,
+ "curve_db": 10.66,
+ "t40_ratio": 1.556,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.0,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.071
+ },
+ "phrase_z": 2.83,
+ "phrase_min": 2.55,
+ "margin": 0.53,
+ "wins": 1.0,
+ "best_confusion": "a wooden stick hit",
+ "as_top": [
+ [
+ "Chopping (food)",
+ 0.649
+ ],
+ [
+ "Finger snapping",
+ 0.025
+ ],
+ [
+ "Chop",
+ 0.024
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.004,
+ "floor_db": -49.8,
+ "stereo_corr": 1.0,
+ "darken": -0.61,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -72.5,
+ "first_sample": 0.0,
+ "tail_db": -56.7,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_loo": [
+ {
+ "file": "E:/earshot/out/refs/click/bench/mk-1124_1.wav",
+ "length_s": 0.055,
+ "typ_z": 0.58,
+ "nn_cos": 0.699,
+ "tone_db": 14.31,
+ "tone_ratio": 1.5,
+ "n_outside": 12,
+ "outside": [
+ [
+ "12.7 kHz",
+ -25.6,
+ "below"
+ ],
+ [
+ "500 Hz",
+ -21.4,
+ "below"
+ ],
+ [
+ "50 Hz",
+ -20.2,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 25.0,
+ "active_s": 0.022,
+ "onset_s": 0.008,
+ "env_oct": 0.972,
+ "curve_db": 13.19,
+ "t40_ratio": 0.455,
+ "attack_ratio": 1.0,
+ "active_ratio": 0.291,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 55.0,
+ "active_s": 0.075
+ },
+ "phrase_z": 2.67,
+ "phrase_min": 2.65,
+ "margin": 0.63,
+ "wins": 1.0,
+ "best_confusion": "a drop of water",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.063
+ ],
+ [
+ "Music",
+ 0.045
+ ],
+ [
+ "Speech",
+ 0.032
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -68.2,
+ "stereo_corr": 1.0,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -94.4,
+ "first_sample": 0.0,
+ "tail_db": -71.2,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/bench/mk-2364_1.wav",
+ "length_s": 0.21,
+ "typ_z": -2.5,
+ "nn_cos": 0.699,
+ "tone_db": 11.83,
+ "tone_ratio": 1.24,
+ "n_outside": 8,
+ "outside": [
+ [
+ "31 Hz",
+ 30.5,
+ "above"
+ ],
+ [
+ "794 Hz",
+ 21.7,
+ "above"
+ ],
+ [
+ "630 Hz",
+ 20.9,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.08,
+ "onset_s": 0.005,
+ "env_oct": 0.205,
+ "curve_db": 17.22,
+ "t40_ratio": 0.857,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.311,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 52.5,
+ "active_s": 0.061
+ },
+ "phrase_z": 2.4,
+ "phrase_min": 2.21,
+ "margin": -0.06,
+ "wins": 0.5,
+ "best_confusion": "a drop of water",
+ "as_top": [
+ [
+ "Plop",
+ 0.667
+ ],
+ [
+ "Music",
+ 0.045
+ ],
+ [
+ "Sound effect",
+ 0.042
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.001,
+ "floor_db": -47.3,
+ "stereo_corr": 1.0,
+ "darken": -2.12,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -53.9,
+ "first_sample": 0.0,
+ "tail_db": -59.7,
+ "lead_s": 0.004,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/bench/fs-610524_1.wav",
+ "length_s": 0.083,
+ "typ_z": -0.31,
+ "nn_cos": 0.712,
+ "tone_db": 11.51,
+ "tone_ratio": 1.21,
+ "n_outside": 9,
+ "outside": [
+ [
+ "99 Hz",
+ 23.1,
+ "above"
+ ],
+ [
+ "198 Hz",
+ 22.8,
+ "above"
+ ],
+ [
+ "125 Hz",
+ 20.4,
+ "above"
+ ]
+ ],
+ "attack_ms": 16.0,
+ "t40_ms": 35.0,
+ "active_s": 0.051,
+ "onset_s": 0.014,
+ "env_oct": 1.739,
+ "curve_db": 12.77,
+ "t40_ratio": 0.636,
+ "attack_ratio": 16.0,
+ "active_ratio": 0.675,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 55.0,
+ "active_s": 0.075
+ },
+ "phrase_z": 2.92,
+ "phrase_min": 2.89,
+ "margin": 1.38,
+ "wins": 1.0,
+ "best_confusion": "a finger snap",
+ "as_top": [
+ [
+ "Speech",
+ 0.053
+ ],
+ [
+ "Music",
+ 0.053
+ ],
+ [
+ "Sound effect",
+ 0.04
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -44.8,
+ "stereo_corr": -0.129,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -61.6,
+ "first_sample": 0.0,
+ "tail_db": -53.6,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/bench/fs-461799_1.wav",
+ "length_s": 0.305,
+ "typ_z": -0.2,
+ "nn_cos": 0.68,
+ "tone_db": 6.79,
+ "tone_ratio": 0.64,
+ "n_outside": 1,
+ "outside": [
+ [
+ "2 kHz",
+ 7.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 149.0,
+ "t40_ms": 110.0,
+ "active_s": 0.255,
+ "onset_s": 0.03,
+ "env_oct": 3.552,
+ "curve_db": 26.17,
+ "t40_ratio": 2.588,
+ "attack_ratio": 149.0,
+ "active_ratio": 4.18,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 42.5,
+ "active_s": 0.061
+ },
+ "phrase_z": 2.34,
+ "phrase_min": 2.33,
+ "margin": -0.44,
+ "wins": 0.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.578
+ ],
+ [
+ "Clapping",
+ 0.447
+ ],
+ [
+ "Speech",
+ 0.045
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.024,
+ "floor_db": -38.9,
+ "stereo_corr": 0.824,
+ "darken": -0.35,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -74.6,
+ "first_sample": 0.0,
+ "tail_db": -48.8,
+ "lead_s": 0.005,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/bench/fs-461799_2.wav",
+ "length_s": 0.109,
+ "typ_z": -0.1,
+ "nn_cos": 0.681,
+ "tone_db": 8.56,
+ "tone_ratio": 0.8,
+ "n_outside": 2,
+ "outside": [
+ [
+ "125 Hz",
+ -21.5,
+ "below"
+ ],
+ [
+ "1.59 kHz",
+ 11.0,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 70.0,
+ "active_s": 0.071,
+ "onset_s": 0.019,
+ "env_oct": 0.279,
+ "curve_db": 12.26,
+ "t40_ratio": 1.647,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.084,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 42.5,
+ "active_s": 0.066
+ },
+ "phrase_z": 2.51,
+ "phrase_min": 2.48,
+ "margin": -0.19,
+ "wins": 0.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Speech",
+ 0.114
+ ],
+ [
+ "Music",
+ 0.034
+ ],
+ [
+ "Single-lens reflex camera",
+ 0.028
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.003,
+ "floor_db": -40.0,
+ "stereo_corr": 0.948,
+ "darken": 0.13,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -90.1,
+ "first_sample": 0.0,
+ "tail_db": -50.3,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/bench/fs-431419_1.wav",
+ "length_s": 0.157,
+ "typ_z": 3.25,
+ "nn_cos": 0.792,
+ "tone_db": 9.19,
+ "tone_ratio": 0.86,
+ "n_outside": 1,
+ "outside": [
+ [
+ "39 Hz",
+ -24.8,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 65.0,
+ "active_s": 0.099,
+ "onset_s": 0.006,
+ "env_oct": 0.437,
+ "curve_db": 14.99,
+ "t40_ratio": 1.529,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.623,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 42.5,
+ "active_s": 0.061
+ },
+ "phrase_z": 2.93,
+ "phrase_min": 2.79,
+ "margin": 0.83,
+ "wins": 1.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Coin (dropping)",
+ 0.274
+ ],
+ [
+ "Chink, clink",
+ 0.077
+ ],
+ [
+ "Speech",
+ 0.069
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.0,
+ "floor_db": -49.6,
+ "stereo_corr": 1.0,
+ "darken": -0.07,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -105.1,
+ "first_sample": 0.0,
+ "tail_db": -60.2,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/bench/fs-431419_2.wav",
+ "length_s": 0.054,
+ "typ_z": 2.85,
+ "nn_cos": 0.792,
+ "tone_db": 9.85,
+ "tone_ratio": 0.95,
+ "n_outside": 3,
+ "outside": [
+ [
+ "10.1 kHz",
+ 15.6,
+ "above"
+ ],
+ [
+ "12.7 kHz",
+ 15.1,
+ "above"
+ ],
+ [
+ "2 kHz",
+ -14.9,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 40.0,
+ "active_s": 0.04,
+ "onset_s": 0.005,
+ "env_oct": 0.459,
+ "curve_db": 12.03,
+ "t40_ratio": 0.727,
+ "attack_ratio": 1.0,
+ "active_ratio": 0.53,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 55.0,
+ "active_s": 0.075
+ },
+ "phrase_z": 3.05,
+ "phrase_min": 2.77,
+ "margin": 0.27,
+ "wins": 0.5,
+ "best_confusion": "a finger snap",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.064
+ ],
+ [
+ "Music",
+ 0.041
+ ],
+ [
+ "Speech",
+ 0.031
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -40.8,
+ "stereo_corr": 1.0,
+ "darken": 0.42,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -89.0,
+ "first_sample": 0.0,
+ "tail_db": -42.6,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_rev": [
+ {
+ "file": "E:/earshot/out/refs/click/bench/mk-1124_1.wav",
+ "length_s": 0.055,
+ "typ_z": -0.27,
+ "nn_cos": 0.755,
+ "tone_db": 16.45,
+ "tone_ratio": 1.77,
+ "n_outside": 1,
+ "outside": [
+ [
+ "62 Hz",
+ -36.9,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 25.0,
+ "active_s": 0.022,
+ "onset_s": 0.008,
+ "env_oct": 0.946,
+ "curve_db": 10.93,
+ "t40_ratio": 0.417,
+ "attack_ratio": 1.0,
+ "active_ratio": 0.336,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 60.0,
+ "active_s": 0.066
+ },
+ "phrase_z": 2.67,
+ "phrase_min": 2.65,
+ "margin": 0.63,
+ "wins": 1.0,
+ "best_confusion": "a drop of water",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.063
+ ],
+ [
+ "Music",
+ 0.045
+ ],
+ [
+ "Speech",
+ 0.032
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -68.2,
+ "stereo_corr": 1.0,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -94.4,
+ "first_sample": 0.0,
+ "tail_db": -71.2,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/bench/mk-2364_1.wav",
+ "length_s": 0.21,
+ "typ_z": -3.09,
+ "nn_cos": 0.604,
+ "tone_db": 7.94,
+ "tone_ratio": 0.85,
+ "n_outside": 3,
+ "outside": [
+ [
+ "630 Hz",
+ 14.3,
+ "above"
+ ],
+ [
+ "794 Hz",
+ 10.9,
+ "above"
+ ],
+ [
+ "500 Hz",
+ 10.7,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 45.0,
+ "active_s": 0.08,
+ "onset_s": 0.005,
+ "env_oct": 0.235,
+ "curve_db": 17.87,
+ "t40_ratio": 0.75,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.221,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 60.0,
+ "active_s": 0.066
+ },
+ "phrase_z": 2.4,
+ "phrase_min": 2.21,
+ "margin": -0.06,
+ "wins": 0.5,
+ "best_confusion": "a drop of water",
+ "as_top": [
+ [
+ "Plop",
+ 0.667
+ ],
+ [
+ "Music",
+ 0.045
+ ],
+ [
+ "Sound effect",
+ 0.042
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.001,
+ "floor_db": -47.3,
+ "stereo_corr": 1.0,
+ "darken": -2.12,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -53.9,
+ "first_sample": 0.0,
+ "tail_db": -59.7,
+ "lead_s": 0.004,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/bench/fs-610524_1.wav",
+ "length_s": 0.083,
+ "typ_z": -0.82,
+ "nn_cos": 0.703,
+ "tone_db": 8.92,
+ "tone_ratio": 0.96,
+ "n_outside": 2,
+ "outside": [
+ [
+ "16 kHz",
+ 26.7,
+ "above"
+ ],
+ [
+ "198 Hz",
+ 16.3,
+ "above"
+ ]
+ ],
+ "attack_ms": 16.0,
+ "t40_ms": 35.0,
+ "active_s": 0.051,
+ "onset_s": 0.014,
+ "env_oct": 1.713,
+ "curve_db": 13.65,
+ "t40_ratio": 0.583,
+ "attack_ratio": 16.0,
+ "active_ratio": 0.779,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 60.0,
+ "active_s": 0.066
+ },
+ "phrase_z": 2.92,
+ "phrase_min": 2.89,
+ "margin": 1.38,
+ "wins": 1.0,
+ "best_confusion": "a finger snap",
+ "as_top": [
+ [
+ "Speech",
+ 0.053
+ ],
+ [
+ "Music",
+ 0.053
+ ],
+ [
+ "Sound effect",
+ 0.04
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -44.8,
+ "stereo_corr": -0.129,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -61.6,
+ "first_sample": 0.0,
+ "tail_db": -53.6,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/bench/fs-461799_1.wav",
+ "length_s": 0.305,
+ "typ_z": -1.52,
+ "nn_cos": 0.708,
+ "tone_db": 7.89,
+ "tone_ratio": 0.85,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 149.0,
+ "t40_ms": 110.0,
+ "active_s": 0.255,
+ "onset_s": 0.03,
+ "env_oct": 3.352,
+ "curve_db": 25.04,
+ "t40_ratio": 1.833,
+ "attack_ratio": 149.0,
+ "active_ratio": 3.893,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 60.0,
+ "active_s": 0.066
+ },
+ "phrase_z": 2.34,
+ "phrase_min": 2.33,
+ "margin": -0.44,
+ "wins": 0.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.578
+ ],
+ [
+ "Clapping",
+ 0.447
+ ],
+ [
+ "Speech",
+ 0.045
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.024,
+ "floor_db": -38.9,
+ "stereo_corr": 0.824,
+ "darken": -0.35,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -74.6,
+ "first_sample": 0.0,
+ "tail_db": -48.8,
+ "lead_s": 0.005,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/bench/fs-461799_2.wav",
+ "length_s": 0.109,
+ "typ_z": -1.73,
+ "nn_cos": 0.713,
+ "tone_db": 9.67,
+ "tone_ratio": 1.04,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 1.0,
+ "t40_ms": 70.0,
+ "active_s": 0.071,
+ "onset_s": 0.019,
+ "env_oct": 0.113,
+ "curve_db": 11.56,
+ "t40_ratio": 1.167,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.084,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 60.0,
+ "active_s": 0.066
+ },
+ "phrase_z": 2.51,
+ "phrase_min": 2.48,
+ "margin": -0.19,
+ "wins": 0.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Speech",
+ 0.114
+ ],
+ [
+ "Music",
+ 0.034
+ ],
+ [
+ "Single-lens reflex camera",
+ 0.028
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.003,
+ "floor_db": -40.0,
+ "stereo_corr": 0.948,
+ "darken": 0.13,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -90.1,
+ "first_sample": 0.0,
+ "tail_db": -50.3,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/bench/fs-431419_1.wav",
+ "length_s": 0.157,
+ "typ_z": -0.02,
+ "nn_cos": 0.737,
+ "tone_db": 11.24,
+ "tone_ratio": 1.21,
+ "n_outside": 2,
+ "outside": [
+ [
+ "16 kHz",
+ 28.7,
+ "above"
+ ],
+ [
+ "39 Hz",
+ -27.2,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 65.0,
+ "active_s": 0.099,
+ "onset_s": 0.006,
+ "env_oct": 0.237,
+ "curve_db": 11.42,
+ "t40_ratio": 1.083,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.511,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 60.0,
+ "active_s": 0.066
+ },
+ "phrase_z": 2.93,
+ "phrase_min": 2.79,
+ "margin": 0.83,
+ "wins": 1.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Coin (dropping)",
+ 0.274
+ ],
+ [
+ "Chink, clink",
+ 0.077
+ ],
+ [
+ "Speech",
+ 0.069
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.0,
+ "floor_db": -49.6,
+ "stereo_corr": 1.0,
+ "darken": -0.07,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -105.1,
+ "first_sample": 0.0,
+ "tail_db": -60.2,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/click/bench/fs-431419_2.wav",
+ "length_s": 0.054,
+ "typ_z": 0.61,
+ "nn_cos": 0.807,
+ "tone_db": 11.37,
+ "tone_ratio": 1.22,
+ "n_outside": 4,
+ "outside": [
+ [
+ "62 Hz",
+ -36.9,
+ "below"
+ ],
+ [
+ "16 kHz",
+ 26.3,
+ "above"
+ ],
+ [
+ "12.7 kHz",
+ 12.6,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 40.0,
+ "active_s": 0.04,
+ "onset_s": 0.005,
+ "env_oct": 0.432,
+ "curve_db": 9.02,
+ "t40_ratio": 0.667,
+ "attack_ratio": 1.0,
+ "active_ratio": 0.611,
+ "ref_median": {
+ "attack_ms": 1.0,
+ "t40_ms": 60.0,
+ "active_s": 0.066
+ },
+ "phrase_z": 3.05,
+ "phrase_min": 2.77,
+ "margin": 0.27,
+ "wins": 0.5,
+ "best_confusion": "a finger snap",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.064
+ ],
+ [
+ "Music",
+ 0.041
+ ],
+ [
+ "Speech",
+ 0.031
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -40.8,
+ "stereo_corr": 1.0,
+ "darken": 0.42,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -89.0,
+ "first_sample": 0.0,
+ "tail_db": -42.6,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity_bench": {
+ "n": 7,
+ "cos_mean": 0.576,
+ "cos_max": 0.792,
+ "tone_min_db": 6.13,
+ "tone_mean_db": 12.82,
+ "near_duplicates": []
+ },
+ "diversity_forge": {
+ "n": 8,
+ "cos_mean": 0.651,
+ "cos_max": 0.92,
+ "tone_min_db": 4.52,
+ "tone_mean_db": 16.52,
+ "near_duplicates": []
+ },
+ "inside": {
+ "per_sound": [
+ 7,
+ 7,
+ 7
+ ],
+ "n_checks": 7,
+ "limits": {
+ "typ_z": -0.508,
+ "tone_db": 13.15,
+ "env_oct": 1.048,
+ "curve_db": 14.422,
+ "phrase_z": 2.488,
+ "margin": -0.086,
+ "as_p": 0.002
+ },
+ "real_per_sound": [
+ 5,
+ 6,
+ 5,
+ 7,
+ 6,
+ 7,
+ 7,
+ 6,
+ 6,
+ 3,
+ 6,
+ 3,
+ 6,
+ 5,
+ 7
+ ]
+ },
+ "summary": {
+ "made": {
+ "n": 3,
+ "typ_z": 1.89,
+ "tone_db": 7.6,
+ "env_oct": 0.56,
+ "curve_db": 8.85,
+ "phrase_z": 3.12,
+ "margin": 1.13,
+ "as_p": 0.0,
+ "tone_ratio": 0.77,
+ "t40_ratio": 0.56,
+ "wins": 1.0,
+ "floor_db": -49.0,
+ "stereo_corr": 1.0,
+ "darken": 0.41,
+ "as_names": 0.0,
+ "artifacts": 0,
+ "var_cos": 0.886
+ },
+ "real_cross": {
+ "n": 8,
+ "typ_z": 0.16,
+ "tone_db": 11.33,
+ "env_oct": 0.22,
+ "curve_db": 9.92,
+ "phrase_z": 2.77,
+ "margin": 0.53,
+ "as_p": 0.0,
+ "tone_ratio": 1.15,
+ "t40_ratio": 1.33,
+ "wins": 0.75,
+ "floor_db": -42.85,
+ "stereo_corr": 0.99,
+ "darken": -0.12,
+ "as_names": 0.12,
+ "artifacts": 0,
+ "var_cos": 0.651
+ },
+ "real_loo": {
+ "n": 7,
+ "typ_z": -0.1,
+ "tone_db": 9.85,
+ "env_oct": 0.46,
+ "curve_db": 13.19,
+ "phrase_z": 2.67,
+ "margin": 0.27,
+ "as_p": 0.0,
+ "tone_ratio": 0.95,
+ "t40_ratio": 0.86,
+ "wins": 0.5,
+ "floor_db": -44.8,
+ "stereo_corr": 1.0,
+ "darken": -0.07,
+ "as_names": 0.0,
+ "artifacts": 0,
+ "var_cos": 0.576
+ }
+ }
+ },
+ "impact": {
+ "n_bench": 1,
+ "n_forge": 2,
+ "note": "thin: the library has three cinematic hits (all three were the v1 impact forge refs); one is held out as bench, so v1 impact bench numbers are optimistic.",
+ "bench_refs": [
+ "E:/earshot/out/refs/impact/bench/kit-impact_03.wav"
+ ],
+ "made": [
+ {
+ "file": "demo/forge-test/impact/final/impact_01.wav",
+ "length_s": 7.014,
+ "nn_cos": 0.769,
+ "tone_db": 6.7,
+ "tone_ratio": null,
+ "n_outside": 19,
+ "outside": [
+ [
+ "16 kHz",
+ -12.6,
+ "below"
+ ],
+ [
+ "794 Hz",
+ -12.2,
+ "below"
+ ],
+ [
+ "1.59 kHz",
+ -9.9,
+ "below"
+ ]
+ ],
+ "attack_ms": 363.0,
+ "t40_ms": 1210.0,
+ "active_s": 4.962,
+ "onset_s": 0.538,
+ "env_oct": 1.568,
+ "curve_db": 24.86,
+ "t40_ratio": 0.351,
+ "attack_ratio": 0.145,
+ "active_ratio": 0.755,
+ "ref_median": {
+ "attack_ms": 2505.0,
+ "t40_ms": 3450.0,
+ "active_s": 6.573
+ },
+ "phrase_z": 3.95,
+ "phrase_min": 3.74,
+ "margin": 1.54,
+ "wins": 1.0,
+ "best_confusion": "a gunshot",
+ "as_top": [
+ [
+ "Explosion",
+ 0.476
+ ],
+ [
+ "Bang",
+ 0.421
+ ],
+ [
+ "Burst, pop",
+ 0.334
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.476,
+ "floor_db": -64.3,
+ "stereo_corr": 0.41,
+ "darken": -1.61,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -137.5,
+ "first_sample": 0.0,
+ "tail_db": -129.5,
+ "lead_s": 0.043,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "demo/forge-test/impact/final/impact_02.wav",
+ "length_s": 7.578,
+ "nn_cos": 0.772,
+ "tone_db": 7.74,
+ "tone_ratio": null,
+ "n_outside": 17,
+ "outside": [
+ [
+ "16 kHz",
+ -17.5,
+ "below"
+ ],
+ [
+ "12.7 kHz",
+ -13.7,
+ "below"
+ ],
+ [
+ "10.1 kHz",
+ -13.4,
+ "below"
+ ]
+ ],
+ "attack_ms": 300.0,
+ "t40_ms": 1740.0,
+ "active_s": 4.413,
+ "onset_s": 0.712,
+ "env_oct": 1.541,
+ "curve_db": 26.47,
+ "t40_ratio": 0.504,
+ "attack_ratio": 0.12,
+ "active_ratio": 0.671,
+ "ref_median": {
+ "attack_ms": 2505.0,
+ "t40_ms": 3450.0,
+ "active_s": 6.573
+ },
+ "phrase_z": 3.68,
+ "phrase_min": 3.54,
+ "margin": 1.81,
+ "wins": 1.0,
+ "best_confusion": "a bass drum",
+ "as_top": [
+ [
+ "Bang",
+ 0.389
+ ],
+ [
+ "Explosion",
+ 0.204
+ ],
+ [
+ "Burst, pop",
+ 0.203
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.389,
+ "floor_db": -69.1,
+ "stereo_corr": 0.183,
+ "darken": -1.11,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -117.1,
+ "first_sample": 0.0,
+ "tail_db": -132.4,
+ "lead_s": 0.049,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "demo/forge-test/impact/final/impact_03.wav",
+ "length_s": 7.23,
+ "nn_cos": 0.773,
+ "tone_db": 6.07,
+ "tone_ratio": null,
+ "n_outside": 20,
+ "outside": [
+ [
+ "16 kHz",
+ -11.2,
+ "below"
+ ],
+ [
+ "1.59 kHz",
+ -10.2,
+ "below"
+ ],
+ [
+ "2 kHz",
+ -10.1,
+ "below"
+ ]
+ ],
+ "attack_ms": 259.0,
+ "t40_ms": 2145.0,
+ "active_s": 4.56,
+ "onset_s": 1.04,
+ "env_oct": 1.496,
+ "curve_db": 23.51,
+ "t40_ratio": 0.622,
+ "attack_ratio": 0.103,
+ "active_ratio": 0.694,
+ "ref_median": {
+ "attack_ms": 2505.0,
+ "t40_ms": 3450.0,
+ "active_s": 6.573
+ },
+ "phrase_z": 3.68,
+ "phrase_min": 3.43,
+ "margin": 1.83,
+ "wins": 1.0,
+ "best_confusion": "a bass drum",
+ "as_top": [
+ [
+ "Explosion",
+ 0.252
+ ],
+ [
+ "Bang",
+ 0.165
+ ],
+ [
+ "Sound effect",
+ 0.159
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.252,
+ "floor_db": -64.7,
+ "stereo_corr": 0.411,
+ "darken": -1.92,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -126.2,
+ "first_sample": 0.0,
+ "tail_db": -126.4,
+ "lead_s": 0.098,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity": {
+ "n": 3,
+ "cos_mean": 0.931,
+ "cos_max": 0.943,
+ "tone_min_db": 3.76,
+ "tone_mean_db": 4.21,
+ "near_duplicates": []
+ },
+ "real_cross": [
+ {
+ "file": "E:/earshot/out/refs/impact/forge/kit-impact_01.wav",
+ "length_s": 4.455,
+ "nn_cos": 0.449,
+ "tone_db": 7.04,
+ "tone_ratio": null,
+ "n_outside": 18,
+ "outside": [
+ [
+ "12.7 kHz",
+ -18.6,
+ "below"
+ ],
+ [
+ "16 kHz",
+ -17.5,
+ "below"
+ ],
+ [
+ "10.1 kHz",
+ -8.5,
+ "below"
+ ]
+ ],
+ "attack_ms": 10.0,
+ "t40_ms": 1310.0,
+ "active_s": 1.812,
+ "onset_s": 0.013,
+ "env_oct": 3.742,
+ "curve_db": 41.06,
+ "t40_ratio": 0.38,
+ "attack_ratio": 0.004,
+ "active_ratio": 0.276,
+ "ref_median": {
+ "attack_ms": 2505.0,
+ "t40_ms": 3450.0,
+ "active_s": 6.573
+ },
+ "phrase_z": 3.04,
+ "phrase_min": 2.25,
+ "margin": 0.69,
+ "wins": 0.5,
+ "best_confusion": "a bass drum",
+ "as_top": [
+ [
+ "Bang",
+ 0.479
+ ],
+ [
+ "Explosion",
+ 0.221
+ ],
+ [
+ "Burst, pop",
+ 0.206
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.479,
+ "floor_db": -88.8,
+ "stereo_corr": 0.338,
+ "darken": -0.56,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -69.1,
+ "first_sample": 0.0,
+ "tail_db": -130.9,
+ "lead_s": 0.007,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/impact/forge/kit-impact_02.wav",
+ "length_s": 6.5,
+ "nn_cos": 0.842,
+ "tone_db": 5.33,
+ "tone_ratio": null,
+ "n_outside": 15,
+ "outside": [
+ [
+ "50 Hz",
+ -13.3,
+ "below"
+ ],
+ [
+ "39 Hz",
+ -10.9,
+ "below"
+ ],
+ [
+ "31 Hz",
+ -8.6,
+ "below"
+ ]
+ ],
+ "attack_ms": 487.0,
+ "t40_ms": 1865.0,
+ "active_s": 4.565,
+ "onset_s": 0.505,
+ "env_oct": 1.259,
+ "curve_db": 22.9,
+ "t40_ratio": 0.541,
+ "attack_ratio": 0.194,
+ "active_ratio": 0.695,
+ "ref_median": {
+ "attack_ms": 2505.0,
+ "t40_ms": 3450.0,
+ "active_s": 6.573
+ },
+ "phrase_z": 3.03,
+ "phrase_min": 1.8,
+ "margin": 1.5,
+ "wins": 0.5,
+ "best_confusion": "a bass drum",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.808
+ ],
+ [
+ "Sound effect",
+ 0.204
+ ],
+ [
+ "Music",
+ 0.062
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.028,
+ "floor_db": -52.3,
+ "stereo_corr": 0.745,
+ "darken": -2.04,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -41.5,
+ "first_sample": 0.0,
+ "tail_db": -100.5,
+ "lead_s": 0.041,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_loo": [],
+ "real_rev": [
+ {
+ "file": "E:/earshot/out/refs/impact/bench/kit-impact_03.wav",
+ "length_s": 7.92,
+ "nn_cos": 0.842,
+ "tone_db": 5.13,
+ "tone_ratio": null,
+ "n_outside": 8,
+ "outside": [
+ [
+ "50 Hz",
+ 10.3,
+ "above"
+ ],
+ [
+ "4 kHz",
+ 7.3,
+ "above"
+ ],
+ [
+ "157 Hz",
+ -7.0,
+ "below"
+ ]
+ ],
+ "attack_ms": 2505.0,
+ "t40_ms": 3450.0,
+ "active_s": 6.573,
+ "onset_s": 0.917,
+ "env_oct": 1.832,
+ "curve_db": 31.74,
+ "t40_ratio": 2.173,
+ "attack_ratio": 10.08,
+ "active_ratio": 2.061,
+ "ref_median": {
+ "attack_ms": 248.5,
+ "t40_ms": 1587.5,
+ "active_s": 3.189
+ },
+ "phrase_z": 2.77,
+ "phrase_min": 1.81,
+ "margin": 1.36,
+ "wins": 0.5,
+ "best_confusion": "a bass drum",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.467
+ ],
+ [
+ "Explosion",
+ 0.337
+ ],
+ [
+ "Burst, pop",
+ 0.135
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.337,
+ "floor_db": -45.2,
+ "stereo_corr": 0.752,
+ "darken": -2.31,
+ "artifacts": [
+ "late start"
+ ],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -68.0,
+ "first_sample": 0.0,
+ "tail_db": -120.0,
+ "lead_s": 0.138,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity_bench": {
+ "n": 1
+ },
+ "diversity_forge": {
+ "n": 2,
+ "cos_mean": 0.659,
+ "cos_max": 0.659,
+ "tone_min_db": 7.13,
+ "tone_mean_db": 7.13,
+ "near_duplicates": []
+ },
+ "inside": {
+ "per_sound": [
+ 5,
+ 5,
+ 6
+ ],
+ "n_checks": 6,
+ "limits": {
+ "tone_db": 6.698,
+ "env_oct": 3.245,
+ "curve_db": 37.428,
+ "phrase_z": 3.032,
+ "margin": 0.852,
+ "as_p": 0.118
+ },
+ "real_per_sound": [
+ 2,
+ 4
+ ]
+ },
+ "summary": {
+ "made": {
+ "n": 3,
+ "tone_db": 6.7,
+ "env_oct": 1.54,
+ "curve_db": 24.86,
+ "phrase_z": 3.68,
+ "margin": 1.81,
+ "as_p": 0.39,
+ "t40_ratio": 0.5,
+ "wins": 1.0,
+ "floor_db": -64.7,
+ "stereo_corr": 0.41,
+ "darken": -1.61,
+ "as_names": 1.0,
+ "artifacts": 0,
+ "var_cos": 0.931
+ },
+ "real_cross": {
+ "n": 2,
+ "tone_db": 6.19,
+ "env_oct": 2.5,
+ "curve_db": 31.98,
+ "phrase_z": 3.04,
+ "margin": 1.09,
+ "as_p": 0.25,
+ "t40_ratio": 0.46,
+ "wins": 0.5,
+ "floor_db": -70.55,
+ "stereo_corr": 0.54,
+ "darken": -1.3,
+ "as_names": 0.5,
+ "artifacts": 0,
+ "var_cos": 0.659
+ },
+ "real_loo": {}
+ }
+ },
+ "pop": {
+ "n_bench": 6,
+ "n_forge": 7,
+ "note": "",
+ "bench_refs": [
+ "E:/earshot/out/refs/pop/bench/fs-202230_1.wav",
+ "E:/earshot/out/refs/pop/bench/fs-33369_1.wav",
+ "E:/earshot/out/refs/pop/bench/fs-545201_1.wav",
+ "E:/earshot/out/refs/pop/bench/fs-573152_1.wav",
+ "E:/earshot/out/refs/pop/bench/mk-3192_1.wav",
+ "E:/earshot/out/refs/pop/bench/mk-3070_1.wav"
+ ],
+ "made": [
+ {
+ "file": "demo/forge-test/pop/final/pop_01.wav",
+ "length_s": 0.304,
+ "typ_z": 1.22,
+ "nn_cos": 0.769,
+ "tone_db": 12.23,
+ "tone_ratio": 0.9,
+ "n_outside": 3,
+ "outside": [
+ [
+ "2 kHz",
+ 24.9,
+ "above"
+ ],
+ [
+ "500 Hz",
+ 13.1,
+ "above"
+ ],
+ [
+ "630 Hz",
+ 12.5,
+ "above"
+ ]
+ ],
+ "attack_ms": 3.0,
+ "t40_ms": 125.0,
+ "active_s": 0.146,
+ "onset_s": 0.029,
+ "env_oct": 0.314,
+ "curve_db": 6.29,
+ "t40_ratio": 1.136,
+ "attack_ratio": 1.5,
+ "active_ratio": 1.127,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 110.0,
+ "active_s": 0.13
+ },
+ "phrase_z": 3.16,
+ "phrase_min": 2.84,
+ "margin": -0.1,
+ "wins": 0.5,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Plop",
+ 0.936
+ ],
+ [
+ "Sound effect",
+ 0.035
+ ],
+ [
+ "Music",
+ 0.031
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.936,
+ "floor_db": -57.1,
+ "stereo_corr": 1.0,
+ "darken": 0.43,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -71.0,
+ "first_sample": 0.0,
+ "tail_db": -65.6,
+ "lead_s": 0.028,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "demo/forge-test/pop/final/pop_02.wav",
+ "length_s": 0.189,
+ "typ_z": 1.67,
+ "nn_cos": 0.826,
+ "tone_db": 8.59,
+ "tone_ratio": 0.63,
+ "n_outside": 2,
+ "outside": [
+ [
+ "630 Hz",
+ 14.9,
+ "above"
+ ],
+ [
+ "99 Hz",
+ 11.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 13.0,
+ "t40_ms": 110.0,
+ "active_s": 0.136,
+ "onset_s": 0.044,
+ "env_oct": 0.924,
+ "curve_db": 7.89,
+ "t40_ratio": 1.0,
+ "attack_ratio": 6.5,
+ "active_ratio": 1.05,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 110.0,
+ "active_s": 0.13
+ },
+ "phrase_z": 3.09,
+ "phrase_min": 2.85,
+ "margin": 0.04,
+ "wins": 0.5,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Plop",
+ 0.884
+ ],
+ [
+ "Music",
+ 0.064
+ ],
+ [
+ "Speech",
+ 0.041
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.884,
+ "floor_db": -68.0,
+ "stereo_corr": 1.0,
+ "darken": 0.19,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -77.8,
+ "first_sample": 0.0,
+ "tail_db": -51.5,
+ "lead_s": 0.041,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "demo/forge-test/pop/final/pop_03.wav",
+ "length_s": 0.204,
+ "typ_z": 1.33,
+ "nn_cos": 0.767,
+ "tone_db": 7.17,
+ "tone_ratio": 0.52,
+ "n_outside": 3,
+ "outside": [
+ [
+ "630 Hz",
+ 14.0,
+ "above"
+ ],
+ [
+ "125 Hz",
+ 13.6,
+ "above"
+ ],
+ [
+ "99 Hz",
+ 13.0,
+ "above"
+ ]
+ ],
+ "attack_ms": 8.0,
+ "t40_ms": 110.0,
+ "active_s": 0.127,
+ "onset_s": 0.038,
+ "env_oct": 0.676,
+ "curve_db": 5.9,
+ "t40_ratio": 1.0,
+ "attack_ratio": 4.0,
+ "active_ratio": 0.981,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 110.0,
+ "active_s": 0.13
+ },
+ "phrase_z": 3.03,
+ "phrase_min": 2.88,
+ "margin": -0.18,
+ "wins": 0.0,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Plop",
+ 0.895
+ ],
+ [
+ "Speech",
+ 0.063
+ ],
+ [
+ "Music",
+ 0.051
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.895,
+ "floor_db": -65.0,
+ "stereo_corr": 1.0,
+ "darken": 0.03,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -87.4,
+ "first_sample": 0.0,
+ "tail_db": -56.6,
+ "lead_s": 0.036,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity": {
+ "n": 3,
+ "cos_mean": 0.917,
+ "cos_max": 0.96,
+ "tone_min_db": 4.49,
+ "tone_mean_db": 8.67,
+ "near_duplicates": []
+ },
+ "real_cross": [
+ {
+ "file": "E:/earshot/out/refs/pop/forge/kit-ui_pop_01.wav",
+ "length_s": 0.19,
+ "typ_z": 2.12,
+ "nn_cos": 0.933,
+ "tone_db": 11.16,
+ "tone_ratio": 0.82,
+ "n_outside": 3,
+ "outside": [
+ [
+ "1.59 kHz",
+ -23.5,
+ "below"
+ ],
+ [
+ "2 kHz",
+ -18.3,
+ "below"
+ ],
+ [
+ "99 Hz",
+ 12.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 6.0,
+ "t40_ms": 35.0,
+ "active_s": 0.101,
+ "onset_s": 0.009,
+ "env_oct": 1.199,
+ "curve_db": 10.63,
+ "t40_ratio": 0.318,
+ "attack_ratio": 3.0,
+ "active_ratio": 0.78,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 110.0,
+ "active_s": 0.13
+ },
+ "phrase_z": 2.81,
+ "phrase_min": 2.55,
+ "margin": -0.91,
+ "wins": 0.0,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Throat clearing",
+ 0.26
+ ],
+ [
+ "Speech",
+ 0.116
+ ],
+ [
+ "Sound effect",
+ 0.065
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.015,
+ "floor_db": -56.0,
+ "stereo_corr": 1.0,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -84.9,
+ "first_sample": 0.0,
+ "tail_db": -58.6,
+ "lead_s": 0.005,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/forge/kit-ui_pop_02.wav",
+ "length_s": 0.3,
+ "typ_z": 0.54,
+ "nn_cos": 0.722,
+ "tone_db": 11.15,
+ "tone_ratio": 0.82,
+ "n_outside": 3,
+ "outside": [
+ [
+ "2.52 kHz",
+ 28.7,
+ "above"
+ ],
+ [
+ "3.17 kHz",
+ 27.6,
+ "above"
+ ],
+ [
+ "2 kHz",
+ 26.8,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 145.0,
+ "active_s": 0.187,
+ "onset_s": 0.008,
+ "env_oct": 0.643,
+ "curve_db": 9.92,
+ "t40_ratio": 1.318,
+ "attack_ratio": 0.5,
+ "active_ratio": 1.444,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 110.0,
+ "active_s": 0.13
+ },
+ "phrase_z": 2.47,
+ "phrase_min": 2.37,
+ "margin": 0.07,
+ "wins": 0.5,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Whip",
+ 0.599
+ ],
+ [
+ "Slap, smack",
+ 0.101
+ ],
+ [
+ "Finger snapping",
+ 0.021
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.004,
+ "floor_db": -46.9,
+ "stereo_corr": 1.0,
+ "darken": 0.04,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -81.7,
+ "first_sample": 0.0,
+ "tail_db": -56.9,
+ "lead_s": 0.005,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/forge/kit-ui_pop_03.wav",
+ "length_s": 0.414,
+ "typ_z": 0.32,
+ "nn_cos": 0.706,
+ "tone_db": 9.4,
+ "tone_ratio": 0.69,
+ "n_outside": 3,
+ "outside": [
+ [
+ "16 kHz",
+ -14.8,
+ "below"
+ ],
+ [
+ "12.7 kHz",
+ -14.4,
+ "below"
+ ],
+ [
+ "500 Hz",
+ 11.8,
+ "above"
+ ]
+ ],
+ "attack_ms": 52.0,
+ "t40_ms": 100.0,
+ "active_s": 0.168,
+ "onset_s": 0.057,
+ "env_oct": 1.738,
+ "curve_db": 26.9,
+ "t40_ratio": 0.909,
+ "attack_ratio": 26.0,
+ "active_ratio": 1.297,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 110.0,
+ "active_s": 0.13
+ },
+ "phrase_z": 2.64,
+ "phrase_min": 2.3,
+ "margin": 0.61,
+ "wins": 0.5,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Plop",
+ 0.487
+ ],
+ [
+ "Sound effect",
+ 0.129
+ ],
+ [
+ "Drip",
+ 0.073
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.487,
+ "floor_db": -62.3,
+ "stereo_corr": 1.0,
+ "darken": 0.38,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -66.0,
+ "first_sample": 0.0,
+ "tail_db": -82.3,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/forge/kit-ui_pop_04.wav",
+ "length_s": 0.236,
+ "typ_z": -1.91,
+ "nn_cos": 0.428,
+ "tone_db": 21.67,
+ "tone_ratio": 1.59,
+ "n_outside": 17,
+ "outside": [
+ [
+ "5.04 kHz",
+ 33.3,
+ "above"
+ ],
+ [
+ "6.35 kHz",
+ 31.3,
+ "above"
+ ],
+ [
+ "198 Hz",
+ 30.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 20.0,
+ "t40_ms": 100.0,
+ "active_s": 0.119,
+ "onset_s": 0.011,
+ "env_oct": 1.194,
+ "curve_db": 9.76,
+ "t40_ratio": 0.909,
+ "attack_ratio": 10.0,
+ "active_ratio": 0.919,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 110.0,
+ "active_s": 0.13
+ },
+ "phrase_z": 0.68,
+ "phrase_min": -0.01,
+ "margin": -1.62,
+ "wins": 0.0,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Chink, clink",
+ 0.059
+ ],
+ [
+ "Clang",
+ 0.048
+ ],
+ [
+ "Dishes, pots, and pans",
+ 0.046
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.004,
+ "floor_db": -92.2,
+ "stereo_corr": 1.0,
+ "darken": -0.35,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -85.5,
+ "first_sample": 0.0,
+ "tail_db": -114.4,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/forge/mk-3000_1.wav",
+ "length_s": 0.5,
+ "typ_z": -0.46,
+ "nn_cos": 0.604,
+ "tone_db": 16.22,
+ "tone_ratio": 1.19,
+ "n_outside": 10,
+ "outside": [
+ [
+ "198 Hz",
+ 31.6,
+ "above"
+ ],
+ [
+ "1 kHz",
+ -28.8,
+ "below"
+ ],
+ [
+ "315 Hz",
+ 24.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 46.0,
+ "t40_ms": 315.0,
+ "active_s": 0.377,
+ "onset_s": 0.008,
+ "env_oct": 2.528,
+ "curve_db": 27.14,
+ "t40_ratio": 2.864,
+ "attack_ratio": 23.0,
+ "active_ratio": 2.911,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 110.0,
+ "active_s": 0.13
+ },
+ "phrase_z": 2.46,
+ "phrase_min": 1.59,
+ "margin": 1.08,
+ "wins": 0.5,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Plop",
+ 0.713
+ ],
+ [
+ "Sound effect",
+ 0.198
+ ],
+ [
+ "Drip",
+ 0.079
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.713,
+ "floor_db": -53.2,
+ "stereo_corr": 0.954,
+ "darken": 1.43,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -84.8,
+ "first_sample": 0.0,
+ "tail_db": -78.6,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/forge/fs-543200_1.wav",
+ "length_s": 0.048,
+ "typ_z": -0.09,
+ "nn_cos": 0.639,
+ "tone_db": 16.06,
+ "tone_ratio": 1.18,
+ "n_outside": 5,
+ "outside": [
+ [
+ "6.35 kHz",
+ 37.5,
+ "above"
+ ],
+ [
+ "5.04 kHz",
+ 34.3,
+ "above"
+ ],
+ [
+ "2.52 kHz",
+ 23.9,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 35.0,
+ "active_s": 0.035,
+ "onset_s": 0.005,
+ "env_oct": 1.513,
+ "curve_db": 20.54,
+ "t40_ratio": 0.318,
+ "attack_ratio": 0.5,
+ "active_ratio": 0.27,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 110.0,
+ "active_s": 0.13
+ },
+ "phrase_z": 1.72,
+ "phrase_min": 1.61,
+ "margin": -0.5,
+ "wins": 0.0,
+ "best_confusion": "a hand clap",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.071
+ ],
+ [
+ "Music",
+ 0.042
+ ],
+ [
+ "Speech",
+ 0.034
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.004,
+ "floor_db": -44.1,
+ "stereo_corr": 1.0,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -64.5,
+ "first_sample": 0.0,
+ "tail_db": -43.2,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/forge/fs-543200_2.wav",
+ "length_s": 0.049,
+ "typ_z": -0.39,
+ "nn_cos": 0.609,
+ "tone_db": 16.06,
+ "tone_ratio": 1.18,
+ "n_outside": 5,
+ "outside": [
+ [
+ "6.35 kHz",
+ 41.4,
+ "above"
+ ],
+ [
+ "5.04 kHz",
+ 32.4,
+ "above"
+ ],
+ [
+ "8 kHz",
+ 32.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 30.0,
+ "active_s": 0.03,
+ "onset_s": 0.005,
+ "env_oct": 1.661,
+ "curve_db": 20.71,
+ "t40_ratio": 0.273,
+ "attack_ratio": 0.5,
+ "active_ratio": 0.232,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 110.0,
+ "active_s": 0.13
+ },
+ "phrase_z": 1.55,
+ "phrase_min": 1.27,
+ "margin": -0.73,
+ "wins": 0.0,
+ "best_confusion": "a hand clap",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.07
+ ],
+ [
+ "Music",
+ 0.043
+ ],
+ [
+ "Speech",
+ 0.034
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.004,
+ "floor_db": -42.4,
+ "stereo_corr": 1.0,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -62.7,
+ "first_sample": 0.0,
+ "tail_db": -46.2,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_loo": [
+ {
+ "file": "E:/earshot/out/refs/pop/bench/fs-202230_1.wav",
+ "length_s": 0.161,
+ "typ_z": -0.66,
+ "nn_cos": 0.642,
+ "tone_db": 14.83,
+ "tone_ratio": 1.1,
+ "n_outside": 7,
+ "outside": [
+ [
+ "157 Hz",
+ -26.6,
+ "below"
+ ],
+ [
+ "62 Hz",
+ -26.2,
+ "below"
+ ],
+ [
+ "79 Hz",
+ -23.7,
+ "below"
+ ]
+ ],
+ "attack_ms": 2.0,
+ "t40_ms": 140.0,
+ "active_s": 0.139,
+ "onset_s": 0.006,
+ "env_oct": 0.213,
+ "curve_db": 13.12,
+ "t40_ratio": 1.4,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.112,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 100.0,
+ "active_s": 0.125
+ },
+ "phrase_z": 1.16,
+ "phrase_min": 1.03,
+ "margin": -1.14,
+ "wins": 0.0,
+ "best_confusion": "a hand clap",
+ "as_top": [
+ [
+ "Plop",
+ 0.159
+ ],
+ [
+ "Boing",
+ 0.131
+ ],
+ [
+ "Frog",
+ 0.102
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.159,
+ "floor_db": -41.9,
+ "stereo_corr": 1.0,
+ "darken": -0.25,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -116.1,
+ "first_sample": 0.0,
+ "tail_db": -48.0,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/bench/fs-33369_1.wav",
+ "length_s": 0.159,
+ "typ_z": 3.39,
+ "nn_cos": 0.832,
+ "tone_db": 13.46,
+ "tone_ratio": 0.97,
+ "n_outside": 5,
+ "outside": [
+ [
+ "1 kHz",
+ -23.5,
+ "below"
+ ],
+ [
+ "39 Hz",
+ 15.5,
+ "above"
+ ],
+ [
+ "157 Hz",
+ 11.4,
+ "above"
+ ]
+ ],
+ "attack_ms": 5.0,
+ "t40_ms": 100.0,
+ "active_s": 0.125,
+ "onset_s": 0.005,
+ "env_oct": 0.562,
+ "curve_db": 13.44,
+ "t40_ratio": 0.833,
+ "attack_ratio": 2.5,
+ "active_ratio": 0.933,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 120.0,
+ "active_s": 0.134
+ },
+ "phrase_z": 2.89,
+ "phrase_min": 2.72,
+ "margin": -0.46,
+ "wins": 0.0,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.141
+ ],
+ [
+ "Speech",
+ 0.094
+ ],
+ [
+ "Plop",
+ 0.085
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.085,
+ "floor_db": -46.5,
+ "stereo_corr": 1.0,
+ "darken": -0.53,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -82.1,
+ "first_sample": 0.0,
+ "tail_db": -55.6,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/bench/fs-545201_1.wav",
+ "length_s": 0.169,
+ "typ_z": 2.92,
+ "nn_cos": 0.816,
+ "tone_db": 13.86,
+ "tone_ratio": 1.03,
+ "n_outside": 7,
+ "outside": [
+ [
+ "1.26 kHz",
+ -32.8,
+ "below"
+ ],
+ [
+ "3.17 kHz",
+ -29.3,
+ "below"
+ ],
+ [
+ "4 kHz",
+ -24.2,
+ "below"
+ ]
+ ],
+ "attack_ms": 2.0,
+ "t40_ms": 125.0,
+ "active_s": 0.145,
+ "onset_s": 0.0,
+ "env_oct": 0.179,
+ "curve_db": 8.51,
+ "t40_ratio": 1.25,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.16,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 100.0,
+ "active_s": 0.125
+ },
+ "phrase_z": 2.8,
+ "phrase_min": 2.54,
+ "margin": 0.17,
+ "wins": 0.5,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Plop",
+ 0.425
+ ],
+ [
+ "Sound effect",
+ 0.102
+ ],
+ [
+ "Speech",
+ 0.078
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.425,
+ "floor_db": -47.9,
+ "stereo_corr": -0.666,
+ "darken": 0.74,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -80.8,
+ "first_sample": 0.0,
+ "tail_db": -60.8,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/bench/fs-573152_1.wav",
+ "length_s": 0.08,
+ "typ_z": 3.55,
+ "nn_cos": 0.832,
+ "tone_db": 11.03,
+ "tone_ratio": 0.8,
+ "n_outside": 3,
+ "outside": [
+ [
+ "5.04 kHz",
+ -23.8,
+ "below"
+ ],
+ [
+ "16 kHz",
+ -11.1,
+ "below"
+ ],
+ [
+ "12.7 kHz",
+ -10.4,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 70.0,
+ "active_s": 0.07,
+ "onset_s": 0.0,
+ "env_oct": 0.905,
+ "curve_db": 16.32,
+ "t40_ratio": 0.583,
+ "attack_ratio": 0.5,
+ "active_ratio": 0.522,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 120.0,
+ "active_s": 0.134
+ },
+ "phrase_z": 2.5,
+ "phrase_min": 2.47,
+ "margin": -0.98,
+ "wins": 0.0,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Speech",
+ 0.177
+ ],
+ [
+ "Throat clearing",
+ 0.057
+ ],
+ [
+ "Music",
+ 0.053
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.003,
+ "floor_db": -37.9,
+ "stereo_corr": 1.0,
+ "darken": 0.3,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -56.0,
+ "first_sample": 0.0,
+ "tail_db": -43.0,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/bench/mk-3192_1.wav",
+ "length_s": 0.215,
+ "typ_z": -0.78,
+ "nn_cos": 0.568,
+ "tone_db": 18.28,
+ "tone_ratio": 1.36,
+ "n_outside": 15,
+ "outside": [
+ [
+ "5.04 kHz",
+ 34.4,
+ "above"
+ ],
+ [
+ "6.35 kHz",
+ 31.5,
+ "above"
+ ],
+ [
+ "630 Hz",
+ -28.2,
+ "below"
+ ]
+ ],
+ "attack_ms": 2.0,
+ "t40_ms": 65.0,
+ "active_s": 0.134,
+ "onset_s": 0.016,
+ "env_oct": 0.328,
+ "curve_db": 11.88,
+ "t40_ratio": 0.542,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.072,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 120.0,
+ "active_s": 0.125
+ },
+ "phrase_z": 1.41,
+ "phrase_min": 0.97,
+ "margin": -1.83,
+ "wins": 0.0,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Speech",
+ 0.05
+ ],
+ [
+ "Plop",
+ 0.05
+ ],
+ [
+ "Coin (dropping)",
+ 0.044
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.05,
+ "floor_db": -52.5,
+ "stereo_corr": 1.0,
+ "darken": 1.71,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -73.3,
+ "first_sample": 0.0,
+ "tail_db": -69.2,
+ "lead_s": 0.006,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/bench/mk-3070_1.wav",
+ "length_s": 0.145,
+ "typ_z": -1.33,
+ "nn_cos": 0.642,
+ "tone_db": 12.85,
+ "tone_ratio": 0.93,
+ "n_outside": 3,
+ "outside": [
+ [
+ "2.52 kHz",
+ 22.7,
+ "above"
+ ],
+ [
+ "2 kHz",
+ 19.2,
+ "above"
+ ],
+ [
+ "315 Hz",
+ 16.5,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 120.0,
+ "active_s": 0.118,
+ "onset_s": 0.002,
+ "env_oct": 0.482,
+ "curve_db": 9.78,
+ "t40_ratio": 1.2,
+ "attack_ratio": 0.5,
+ "active_ratio": 0.881,
+ "ref_median": {
+ "attack_ms": 2.0,
+ "t40_ms": 100.0,
+ "active_s": 0.134
+ },
+ "phrase_z": 1.21,
+ "phrase_min": 0.86,
+ "margin": -1.6,
+ "wins": 0.0,
+ "best_confusion": "a hand clap",
+ "as_top": [
+ [
+ "Gasp",
+ 0.135
+ ],
+ [
+ "Speech",
+ 0.039
+ ],
+ [
+ "Scrape",
+ 0.037
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.006,
+ "floor_db": -56.7,
+ "stereo_corr": 0.87,
+ "darken": -0.05,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -76.2,
+ "first_sample": 0.0,
+ "tail_db": -77.0,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_rev": [
+ {
+ "file": "E:/earshot/out/refs/pop/bench/fs-202230_1.wav",
+ "length_s": 0.161,
+ "typ_z": -1.02,
+ "nn_cos": 0.559,
+ "tone_db": 13.87,
+ "tone_ratio": 0.89,
+ "n_outside": 9,
+ "outside": [
+ [
+ "157 Hz",
+ -29.4,
+ "below"
+ ],
+ [
+ "125 Hz",
+ -28.6,
+ "below"
+ ],
+ [
+ "99 Hz",
+ -22.9,
+ "below"
+ ]
+ ],
+ "attack_ms": 2.0,
+ "t40_ms": 140.0,
+ "active_s": 0.139,
+ "onset_s": 0.006,
+ "env_oct": 0.765,
+ "curve_db": 6.72,
+ "t40_ratio": 1.4,
+ "attack_ratio": 0.333,
+ "active_ratio": 1.168,
+ "ref_median": {
+ "attack_ms": 6.0,
+ "t40_ms": 100.0,
+ "active_s": 0.119
+ },
+ "phrase_z": 1.16,
+ "phrase_min": 1.03,
+ "margin": -1.14,
+ "wins": 0.0,
+ "best_confusion": "a hand clap",
+ "as_top": [
+ [
+ "Plop",
+ 0.159
+ ],
+ [
+ "Boing",
+ 0.131
+ ],
+ [
+ "Frog",
+ 0.102
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.159,
+ "floor_db": -41.9,
+ "stereo_corr": 1.0,
+ "darken": -0.25,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -116.1,
+ "first_sample": 0.0,
+ "tail_db": -48.0,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/bench/fs-33369_1.wav",
+ "length_s": 0.159,
+ "typ_z": 1.53,
+ "nn_cos": 0.933,
+ "tone_db": 12.12,
+ "tone_ratio": 0.78,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 5.0,
+ "t40_ms": 100.0,
+ "active_s": 0.125,
+ "onset_s": 0.005,
+ "env_oct": 0.111,
+ "curve_db": 3.86,
+ "t40_ratio": 1.0,
+ "attack_ratio": 0.833,
+ "active_ratio": 1.05,
+ "ref_median": {
+ "attack_ms": 6.0,
+ "t40_ms": 100.0,
+ "active_s": 0.119
+ },
+ "phrase_z": 2.89,
+ "phrase_min": 2.72,
+ "margin": -0.46,
+ "wins": 0.0,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.141
+ ],
+ [
+ "Speech",
+ 0.094
+ ],
+ [
+ "Plop",
+ 0.085
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.085,
+ "floor_db": -46.5,
+ "stereo_corr": 1.0,
+ "darken": -0.53,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -82.1,
+ "first_sample": 0.0,
+ "tail_db": -55.6,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/bench/fs-545201_1.wav",
+ "length_s": 0.169,
+ "typ_z": 1.02,
+ "nn_cos": 0.81,
+ "tone_db": 15.15,
+ "tone_ratio": 0.97,
+ "n_outside": 3,
+ "outside": [
+ [
+ "3.17 kHz",
+ -37.4,
+ "below"
+ ],
+ [
+ "2.52 kHz",
+ -32.8,
+ "below"
+ ],
+ [
+ "397 Hz",
+ -6.5,
+ "below"
+ ]
+ ],
+ "attack_ms": 2.0,
+ "t40_ms": 125.0,
+ "active_s": 0.145,
+ "onset_s": 0.0,
+ "env_oct": 0.731,
+ "curve_db": 7.07,
+ "t40_ratio": 1.25,
+ "attack_ratio": 0.333,
+ "active_ratio": 1.218,
+ "ref_median": {
+ "attack_ms": 6.0,
+ "t40_ms": 100.0,
+ "active_s": 0.119
+ },
+ "phrase_z": 2.8,
+ "phrase_min": 2.54,
+ "margin": 0.17,
+ "wins": 0.5,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Plop",
+ 0.425
+ ],
+ [
+ "Sound effect",
+ 0.102
+ ],
+ [
+ "Speech",
+ 0.078
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.425,
+ "floor_db": -47.9,
+ "stereo_corr": -0.666,
+ "darken": 0.74,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -80.8,
+ "first_sample": 0.0,
+ "tail_db": -60.8,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/bench/fs-573152_1.wav",
+ "length_s": 0.08,
+ "typ_z": 0.98,
+ "nn_cos": 0.789,
+ "tone_db": 13.94,
+ "tone_ratio": 0.9,
+ "n_outside": 5,
+ "outside": [
+ [
+ "1.26 kHz",
+ 20.4,
+ "above"
+ ],
+ [
+ "5.04 kHz",
+ -16.8,
+ "below"
+ ],
+ [
+ "397 Hz",
+ -11.3,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 70.0,
+ "active_s": 0.07,
+ "onset_s": 0.0,
+ "env_oct": 1.288,
+ "curve_db": 9.79,
+ "t40_ratio": 0.7,
+ "attack_ratio": 0.167,
+ "active_ratio": 0.588,
+ "ref_median": {
+ "attack_ms": 6.0,
+ "t40_ms": 100.0,
+ "active_s": 0.119
+ },
+ "phrase_z": 2.5,
+ "phrase_min": 2.47,
+ "margin": -0.98,
+ "wins": 0.0,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Speech",
+ 0.177
+ ],
+ [
+ "Throat clearing",
+ 0.057
+ ],
+ [
+ "Music",
+ 0.053
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.003,
+ "floor_db": -37.9,
+ "stereo_corr": 1.0,
+ "darken": 0.3,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -56.0,
+ "first_sample": 0.0,
+ "tail_db": -43.0,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/bench/mk-3192_1.wav",
+ "length_s": 0.215,
+ "typ_z": -0.49,
+ "nn_cos": 0.706,
+ "tone_db": 16.77,
+ "tone_ratio": 1.08,
+ "n_outside": 11,
+ "outside": [
+ [
+ "315 Hz",
+ -24.7,
+ "below"
+ ],
+ [
+ "630 Hz",
+ -22.9,
+ "below"
+ ],
+ [
+ "250 Hz",
+ -22.8,
+ "below"
+ ]
+ ],
+ "attack_ms": 2.0,
+ "t40_ms": 65.0,
+ "active_s": 0.134,
+ "onset_s": 0.016,
+ "env_oct": 0.793,
+ "curve_db": 7.79,
+ "t40_ratio": 0.65,
+ "attack_ratio": 0.333,
+ "active_ratio": 1.126,
+ "ref_median": {
+ "attack_ms": 6.0,
+ "t40_ms": 100.0,
+ "active_s": 0.119
+ },
+ "phrase_z": 1.41,
+ "phrase_min": 0.97,
+ "margin": -1.83,
+ "wins": 0.0,
+ "best_confusion": "a tongue click",
+ "as_top": [
+ [
+ "Speech",
+ 0.05
+ ],
+ [
+ "Plop",
+ 0.05
+ ],
+ [
+ "Coin (dropping)",
+ 0.044
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.05,
+ "floor_db": -52.5,
+ "stereo_corr": 1.0,
+ "darken": 1.71,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -73.3,
+ "first_sample": 0.0,
+ "tail_db": -69.2,
+ "lead_s": 0.006,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/pop/bench/mk-3070_1.wav",
+ "length_s": 0.145,
+ "typ_z": -0.8,
+ "nn_cos": 0.604,
+ "tone_db": 7.28,
+ "tone_ratio": 0.47,
+ "n_outside": 1,
+ "outside": [
+ [
+ "1.26 kHz",
+ 14.9,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 120.0,
+ "active_s": 0.118,
+ "onset_s": 0.002,
+ "env_oct": 0.953,
+ "curve_db": 8.3,
+ "t40_ratio": 1.2,
+ "attack_ratio": 0.167,
+ "active_ratio": 0.992,
+ "ref_median": {
+ "attack_ms": 6.0,
+ "t40_ms": 100.0,
+ "active_s": 0.119
+ },
+ "phrase_z": 1.21,
+ "phrase_min": 0.86,
+ "margin": -1.6,
+ "wins": 0.0,
+ "best_confusion": "a hand clap",
+ "as_top": [
+ [
+ "Gasp",
+ 0.135
+ ],
+ [
+ "Speech",
+ 0.039
+ ],
+ [
+ "Scrape",
+ 0.037
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.006,
+ "floor_db": -56.7,
+ "stereo_corr": 0.87,
+ "darken": -0.05,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -76.2,
+ "first_sample": 0.0,
+ "tail_db": -77.0,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity_bench": {
+ "n": 6,
+ "cos_mean": 0.522,
+ "cos_max": 0.832,
+ "tone_min_db": 7.62,
+ "tone_mean_db": 16.34,
+ "near_duplicates": []
+ },
+ "diversity_forge": {
+ "n": 7,
+ "cos_mean": 0.501,
+ "cos_max": 0.978,
+ "tone_min_db": 3.32,
+ "tone_mean_db": 18.73,
+ "near_duplicates": []
+ },
+ "inside": {
+ "per_sound": [
+ 7,
+ 7,
+ 7
+ ],
+ "n_checks": 7,
+ "limits": {
+ "typ_z": -0.732,
+ "tone_db": 16.156,
+ "env_oct": 1.602,
+ "curve_db": 20.642,
+ "phrase_z": 1.29,
+ "margin": -1.416,
+ "as_p": 0.004
+ },
+ "real_per_sound": [
+ 7,
+ 7,
+ 5,
+ 3,
+ 4,
+ 7,
+ 5,
+ 6,
+ 7,
+ 7,
+ 6,
+ 4,
+ 4
+ ]
+ },
+ "summary": {
+ "made": {
+ "n": 3,
+ "typ_z": 1.33,
+ "tone_db": 8.59,
+ "env_oct": 0.68,
+ "curve_db": 6.29,
+ "phrase_z": 3.09,
+ "margin": -0.1,
+ "as_p": 0.9,
+ "tone_ratio": 0.63,
+ "t40_ratio": 1.0,
+ "wins": 0.5,
+ "floor_db": -65.0,
+ "stereo_corr": 1.0,
+ "darken": 0.19,
+ "as_names": 1.0,
+ "artifacts": 0,
+ "var_cos": 0.917
+ },
+ "real_cross": {
+ "n": 7,
+ "typ_z": -0.09,
+ "tone_db": 16.06,
+ "env_oct": 1.51,
+ "curve_db": 20.54,
+ "phrase_z": 2.46,
+ "margin": -0.5,
+ "as_p": 0.0,
+ "tone_ratio": 1.18,
+ "t40_ratio": 0.91,
+ "wins": 0.0,
+ "floor_db": -53.2,
+ "stereo_corr": 1.0,
+ "darken": 0.21,
+ "as_names": 0.29,
+ "artifacts": 0,
+ "var_cos": 0.501
+ },
+ "real_loo": {
+ "n": 6,
+ "typ_z": 1.13,
+ "tone_db": 13.66,
+ "env_oct": 0.41,
+ "curve_db": 12.5,
+ "phrase_z": 1.96,
+ "margin": -1.06,
+ "as_p": 0.07,
+ "tone_ratio": 1.0,
+ "t40_ratio": 1.02,
+ "wins": 0.0,
+ "floor_db": -47.2,
+ "stereo_corr": 1.0,
+ "darken": 0.12,
+ "as_names": 0.67,
+ "artifacts": 0,
+ "var_cos": 0.522
+ }
+ }
+ },
+ "typing_key": {
+ "n_bench": 12,
+ "n_forge": 15,
+ "note": "",
+ "bench_refs": [
+ "E:/earshot/out/refs/typing_key/bench/fs-378083_1.wav",
+ "E:/earshot/out/refs/typing_key/bench/fs-378084_1.wav",
+ "E:/earshot/out/refs/typing_key/bench/fs-378085_1.wav",
+ "E:/earshot/out/refs/typing_key/bench/fs-570754_1.wav",
+ "E:/earshot/out/refs/typing_key/bench/fs-766639_1.wav",
+ "E:/earshot/out/refs/typing_key/bench/fs-766640_1.wav",
+ "E:/earshot/out/refs/typing_key/bench/fs-609534_1.wav",
+ "E:/earshot/out/refs/typing_key/bench/fs-609534_2.wav",
+ "E:/earshot/out/refs/typing_key/bench/fs-609534_3.wav",
+ "E:/earshot/out/refs/typing_key/bench/fs-515810_1.wav",
+ "E:/earshot/out/refs/typing_key/bench/fs-815614_1.wav",
+ "E:/earshot/out/refs/typing_key/bench/fs-815614_2.wav"
+ ],
+ "made": [
+ {
+ "file": "demo/forge-test/keypress/final/keypress_01.wav",
+ "length_s": 0.422,
+ "typ_z": 1.27,
+ "nn_cos": 0.916,
+ "tone_db": 8.88,
+ "tone_ratio": 1.62,
+ "n_outside": 5,
+ "outside": [
+ [
+ "31 Hz",
+ -20.8,
+ "below"
+ ],
+ [
+ "79 Hz",
+ 18.8,
+ "above"
+ ],
+ [
+ "794 Hz",
+ -15.9,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 220.0,
+ "active_s": 0.249,
+ "onset_s": 0.026,
+ "env_oct": 1.128,
+ "curve_db": 18.27,
+ "t40_ratio": 1.833,
+ "attack_ratio": 0.2,
+ "active_ratio": 1.14,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 5.28,
+ "phrase_min": 4.92,
+ "margin": 3.2,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Knock",
+ 0.322
+ ],
+ [
+ "Sound effect",
+ 0.093
+ ],
+ [
+ "Tap",
+ 0.091
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.065,
+ "floor_db": -47.7,
+ "stereo_corr": 0.937,
+ "darken": 0.75,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -120.1,
+ "first_sample": 0.0,
+ "tail_db": -70.9,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "demo/forge-test/keypress/final/keypress_02.wav",
+ "length_s": 0.377,
+ "typ_z": 1.2,
+ "nn_cos": 0.924,
+ "tone_db": 6.75,
+ "tone_ratio": 1.23,
+ "n_outside": 3,
+ "outside": [
+ [
+ "79 Hz",
+ 17.2,
+ "above"
+ ],
+ [
+ "31 Hz",
+ -15.9,
+ "below"
+ ],
+ [
+ "1.26 kHz",
+ -7.9,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 155.0,
+ "active_s": 0.176,
+ "onset_s": 0.019,
+ "env_oct": 1.001,
+ "curve_db": 11.05,
+ "t40_ratio": 1.292,
+ "attack_ratio": 0.2,
+ "active_ratio": 0.805,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 5.67,
+ "phrase_min": 5.25,
+ "margin": 3.74,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Knock",
+ 0.095
+ ],
+ [
+ "Sound effect",
+ 0.076
+ ],
+ [
+ "Coin (dropping)",
+ 0.059
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.027,
+ "floor_db": -54.7,
+ "stereo_corr": 0.878,
+ "darken": -0.8,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -123.6,
+ "first_sample": 0.0,
+ "tail_db": -78.2,
+ "lead_s": 0.014,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "demo/forge-test/keypress/final/keypress_03.wav",
+ "length_s": 0.388,
+ "typ_z": 0.91,
+ "nn_cos": 0.881,
+ "tone_db": 8.2,
+ "tone_ratio": 1.5,
+ "n_outside": 7,
+ "outside": [
+ [
+ "79 Hz",
+ 20.7,
+ "above"
+ ],
+ [
+ "31 Hz",
+ -20.1,
+ "below"
+ ],
+ [
+ "5.04 kHz",
+ -13.2,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 200.0,
+ "active_s": 0.197,
+ "onset_s": 0.023,
+ "env_oct": 1.069,
+ "curve_db": 14.32,
+ "t40_ratio": 1.667,
+ "attack_ratio": 0.2,
+ "active_ratio": 0.902,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 5.56,
+ "phrase_min": 5.07,
+ "margin": 3.57,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Computer keyboard",
+ 0.14
+ ],
+ [
+ "Sound effect",
+ 0.118
+ ],
+ [
+ "Typing",
+ 0.111
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.14,
+ "floor_db": -55.3,
+ "stereo_corr": 0.958,
+ "darken": 0.4,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -147.2,
+ "first_sample": 0.0,
+ "tail_db": -81.3,
+ "lead_s": 0.017,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity": {
+ "n": 3,
+ "cos_mean": 0.965,
+ "cos_max": 0.967,
+ "tone_min_db": 4.27,
+ "tone_mean_db": 4.93,
+ "near_duplicates": []
+ },
+ "real_cross": [
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/kit-kb_thock_01.wav",
+ "length_s": 0.254,
+ "typ_z": 1.34,
+ "nn_cos": 0.926,
+ "tone_db": 9.89,
+ "tone_ratio": 1.81,
+ "n_outside": 7,
+ "outside": [
+ [
+ "31 Hz",
+ -17.6,
+ "below"
+ ],
+ [
+ "630 Hz",
+ -17.3,
+ "below"
+ ],
+ [
+ "794 Hz",
+ -16.2,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 205.0,
+ "active_s": 0.218,
+ "onset_s": 0.012,
+ "env_oct": 1.033,
+ "curve_db": 15.21,
+ "t40_ratio": 1.708,
+ "attack_ratio": 0.2,
+ "active_ratio": 0.998,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 5.48,
+ "phrase_min": 5.14,
+ "margin": 3.64,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.306
+ ],
+ [
+ "Computer keyboard",
+ 0.076
+ ],
+ [
+ "Tap",
+ 0.058
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.076,
+ "floor_db": -40.0,
+ "stereo_corr": 0.668,
+ "darken": -0.39,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -97.7,
+ "first_sample": 0.0,
+ "tail_db": -53.1,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/kit-kb_thock_02.wav",
+ "length_s": 0.281,
+ "typ_z": 0.98,
+ "nn_cos": 0.905,
+ "tone_db": 10.24,
+ "tone_ratio": 1.87,
+ "n_outside": 5,
+ "outside": [
+ [
+ "630 Hz",
+ -23.9,
+ "below"
+ ],
+ [
+ "62 Hz",
+ 17.4,
+ "above"
+ ],
+ [
+ "99 Hz",
+ 15.3,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 175.0,
+ "active_s": 0.255,
+ "onset_s": 0.005,
+ "env_oct": 1.03,
+ "curve_db": 19.49,
+ "t40_ratio": 1.458,
+ "attack_ratio": 0.2,
+ "active_ratio": 1.167,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 5.18,
+ "phrase_min": 4.93,
+ "margin": 3.4,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Computer keyboard",
+ 0.481
+ ],
+ [
+ "Typing",
+ 0.285
+ ],
+ [
+ "Sound effect",
+ 0.05
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.481,
+ "floor_db": -38.9,
+ "stereo_corr": 0.656,
+ "darken": 0.13,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -95.8,
+ "first_sample": 0.0,
+ "tail_db": -61.0,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/kit-kb_thock_03.wav",
+ "length_s": 0.362,
+ "typ_z": 1.25,
+ "nn_cos": 0.927,
+ "tone_db": 9.38,
+ "tone_ratio": 1.71,
+ "n_outside": 4,
+ "outside": [
+ [
+ "630 Hz",
+ -17.6,
+ "below"
+ ],
+ [
+ "62 Hz",
+ 15.8,
+ "above"
+ ],
+ [
+ "500 Hz",
+ -14.0,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 185.0,
+ "active_s": 0.269,
+ "onset_s": 0.041,
+ "env_oct": 1.082,
+ "curve_db": 14.68,
+ "t40_ratio": 1.542,
+ "attack_ratio": 0.2,
+ "active_ratio": 1.231,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 5.42,
+ "phrase_min": 4.97,
+ "margin": 3.87,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Computer keyboard",
+ 0.528
+ ],
+ [
+ "Typing",
+ 0.355
+ ],
+ [
+ "Sound effect",
+ 0.076
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.528,
+ "floor_db": -51.2,
+ "stereo_corr": 0.591,
+ "darken": 0.01,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -96.7,
+ "first_sample": 0.0,
+ "tail_db": -59.9,
+ "lead_s": 0.005,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/kit-kb_thock_04.wav",
+ "length_s": 0.234,
+ "typ_z": 1.35,
+ "nn_cos": 0.908,
+ "tone_db": 7.72,
+ "tone_ratio": 1.41,
+ "n_outside": 2,
+ "outside": [
+ [
+ "62 Hz",
+ 14.9,
+ "above"
+ ],
+ [
+ "500 Hz",
+ -9.5,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 175.0,
+ "active_s": 0.173,
+ "onset_s": 0.042,
+ "env_oct": 1.068,
+ "curve_db": 9.55,
+ "t40_ratio": 1.458,
+ "attack_ratio": 0.2,
+ "active_ratio": 0.792,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 4.75,
+ "phrase_min": 4.42,
+ "margin": 2.83,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Computer keyboard",
+ 0.077
+ ],
+ [
+ "Sound effect",
+ 0.063
+ ],
+ [
+ "Typing",
+ 0.044
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.077,
+ "floor_db": -43.7,
+ "stereo_corr": 0.528,
+ "darken": -0.8,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -84.1,
+ "first_sample": 0.0,
+ "tail_db": -61.8,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/kit-kb_enter_01.wav",
+ "length_s": 0.4,
+ "typ_z": 1.26,
+ "nn_cos": 0.917,
+ "tone_db": 8.82,
+ "tone_ratio": 1.61,
+ "n_outside": 5,
+ "outside": [
+ [
+ "62 Hz",
+ 18.7,
+ "above"
+ ],
+ [
+ "5.04 kHz",
+ -16.7,
+ "below"
+ ],
+ [
+ "4 kHz",
+ -14.3,
+ "below"
+ ]
+ ],
+ "attack_ms": 3.0,
+ "t40_ms": 200.0,
+ "active_s": 0.274,
+ "onset_s": 0.016,
+ "env_oct": 0.6,
+ "curve_db": 19.85,
+ "t40_ratio": 1.667,
+ "attack_ratio": 0.6,
+ "active_ratio": 1.254,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 4.89,
+ "phrase_min": 4.43,
+ "margin": 3.05,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.167
+ ],
+ [
+ "Tap",
+ 0.064
+ ],
+ [
+ "Sound effect",
+ 0.062
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.05,
+ "floor_db": -53.9,
+ "stereo_corr": 0.76,
+ "darken": -0.83,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -89.8,
+ "first_sample": 0.0,
+ "tail_db": -80.9,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/kit-kb_space_01.wav",
+ "length_s": 0.33,
+ "typ_z": 0.01,
+ "nn_cos": 0.813,
+ "tone_db": 7.12,
+ "tone_ratio": 1.3,
+ "n_outside": 2,
+ "outside": [
+ [
+ "62 Hz",
+ 17.8,
+ "above"
+ ],
+ [
+ "79 Hz",
+ 14.0,
+ "above"
+ ]
+ ],
+ "attack_ms": 2.0,
+ "t40_ms": 160.0,
+ "active_s": 0.228,
+ "onset_s": 0.017,
+ "env_oct": 0.599,
+ "curve_db": 16.65,
+ "t40_ratio": 1.333,
+ "attack_ratio": 0.4,
+ "active_ratio": 1.043,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 5.25,
+ "phrase_min": 4.88,
+ "margin": 4.11,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.084
+ ],
+ [
+ "Crack",
+ 0.043
+ ],
+ [
+ "Creak",
+ 0.036
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.012,
+ "floor_db": -61.4,
+ "stereo_corr": 0.578,
+ "darken": 0.22,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -90.1,
+ "first_sample": 0.0,
+ "tail_db": -86.0,
+ "lead_s": 0.005,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/kit-kb_click_01.wav",
+ "length_s": 0.254,
+ "typ_z": -0.47,
+ "nn_cos": 0.812,
+ "tone_db": 14.97,
+ "tone_ratio": 2.73,
+ "n_outside": 7,
+ "outside": [
+ [
+ "62 Hz",
+ -37.7,
+ "below"
+ ],
+ [
+ "50 Hz",
+ -32.5,
+ "below"
+ ],
+ [
+ "79 Hz",
+ -29.5,
+ "below"
+ ]
+ ],
+ "attack_ms": 15.0,
+ "t40_ms": 120.0,
+ "active_s": 0.134,
+ "onset_s": 0.006,
+ "env_oct": 0.763,
+ "curve_db": 9.52,
+ "t40_ratio": 1.0,
+ "attack_ratio": 3.0,
+ "active_ratio": 0.613,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 4.4,
+ "phrase_min": 3.68,
+ "margin": 2.21,
+ "wins": 1.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Single-lens reflex camera",
+ 0.069
+ ],
+ [
+ "Coin (dropping)",
+ 0.038
+ ],
+ [
+ "Sound effect",
+ 0.03
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.008,
+ "floor_db": -59.7,
+ "stereo_corr": 1.0,
+ "darken": 0.66,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -84.8,
+ "first_sample": 0.0,
+ "tail_db": -74.1,
+ "lead_s": 0.005,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/kit-kb_click_02.wav",
+ "length_s": 0.284,
+ "typ_z": -0.29,
+ "nn_cos": 0.802,
+ "tone_db": 14.0,
+ "tone_ratio": 2.56,
+ "n_outside": 6,
+ "outside": [
+ [
+ "79 Hz",
+ -34.9,
+ "below"
+ ],
+ [
+ "62 Hz",
+ -30.8,
+ "below"
+ ],
+ [
+ "50 Hz",
+ -25.8,
+ "below"
+ ]
+ ],
+ "attack_ms": 3.0,
+ "t40_ms": 110.0,
+ "active_s": 0.109,
+ "onset_s": 0.011,
+ "env_oct": 0.622,
+ "curve_db": 9.78,
+ "t40_ratio": 0.917,
+ "attack_ratio": 0.6,
+ "active_ratio": 0.499,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 4.69,
+ "phrase_min": 4.18,
+ "margin": 1.99,
+ "wins": 1.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Single-lens reflex camera",
+ 0.069
+ ],
+ [
+ "Coin (dropping)",
+ 0.055
+ ],
+ [
+ "Sound effect",
+ 0.04
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.012,
+ "floor_db": -67.6,
+ "stereo_corr": 1.0,
+ "darken": 0.72,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -79.1,
+ "first_sample": 0.0,
+ "tail_db": -88.7,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/kit-kb_enter_02.wav",
+ "length_s": 0.18,
+ "typ_z": -0.43,
+ "nn_cos": 0.783,
+ "tone_db": 8.88,
+ "tone_ratio": 1.62,
+ "n_outside": 1,
+ "outside": [
+ [
+ "16 kHz",
+ 20.8,
+ "above"
+ ]
+ ],
+ "attack_ms": 10.0,
+ "t40_ms": 155.0,
+ "active_s": 0.164,
+ "onset_s": 0.006,
+ "env_oct": 0.594,
+ "curve_db": 9.68,
+ "t40_ratio": 1.292,
+ "attack_ratio": 2.0,
+ "active_ratio": 0.751,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 3.33,
+ "phrase_min": 3.15,
+ "margin": 1.24,
+ "wins": 1.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Typewriter",
+ 0.088
+ ],
+ [
+ "Computer keyboard",
+ 0.072
+ ],
+ [
+ "Speech",
+ 0.056
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.088,
+ "floor_db": -36.4,
+ "stereo_corr": 1.0,
+ "darken": -0.5,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -95.9,
+ "first_sample": 0.0,
+ "tail_db": -46.3,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/fs-561697_1.wav",
+ "length_s": 0.18,
+ "typ_z": 0.26,
+ "nn_cos": 0.845,
+ "tone_db": 15.76,
+ "tone_ratio": 2.88,
+ "n_outside": 8,
+ "outside": [
+ [
+ "50 Hz",
+ -35.2,
+ "below"
+ ],
+ [
+ "79 Hz",
+ -32.8,
+ "below"
+ ],
+ [
+ "62 Hz",
+ -30.3,
+ "below"
+ ]
+ ],
+ "attack_ms": 17.0,
+ "t40_ms": 140.0,
+ "active_s": 0.155,
+ "onset_s": 0.005,
+ "env_oct": 0.828,
+ "curve_db": 17.09,
+ "t40_ratio": 1.167,
+ "attack_ratio": 3.4,
+ "active_ratio": 0.709,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 4.18,
+ "phrase_min": 3.68,
+ "margin": 1.42,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Single-lens reflex camera",
+ 0.34
+ ],
+ [
+ "Camera",
+ 0.045
+ ],
+ [
+ "Sound effect",
+ 0.043
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.012,
+ "floor_db": -40.4,
+ "stereo_corr": 1.0,
+ "darken": 0.04,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -73.5,
+ "first_sample": 0.0,
+ "tail_db": -56.6,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/fs-561699_1.wav",
+ "length_s": 0.419,
+ "typ_z": -0.98,
+ "nn_cos": 0.717,
+ "tone_db": 16.54,
+ "tone_ratio": 3.02,
+ "n_outside": 9,
+ "outside": [
+ [
+ "62 Hz",
+ -34.6,
+ "below"
+ ],
+ [
+ "79 Hz",
+ -33.1,
+ "below"
+ ],
+ [
+ "99 Hz",
+ -31.4,
+ "below"
+ ]
+ ],
+ "attack_ms": 17.0,
+ "t40_ms": 80.0,
+ "active_s": 0.094,
+ "onset_s": 0.021,
+ "env_oct": 1.189,
+ "curve_db": 8.6,
+ "t40_ratio": 0.667,
+ "attack_ratio": 3.4,
+ "active_ratio": 0.43,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 2.81,
+ "phrase_min": 2.37,
+ "margin": -0.23,
+ "wins": 0.5,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.054
+ ],
+ [
+ "Single-lens reflex camera",
+ 0.048
+ ],
+ [
+ "Coin (dropping)",
+ 0.028
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.002,
+ "floor_db": -62.9,
+ "stereo_corr": 1.0,
+ "darken": -0.66,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -85.3,
+ "first_sample": 0.0,
+ "tail_db": -68.9,
+ "lead_s": 0.004,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/fs-709460_1.wav",
+ "length_s": 0.278,
+ "typ_z": -0.15,
+ "nn_cos": 0.841,
+ "tone_db": 4.88,
+ "tone_ratio": 0.89,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 35.0,
+ "t40_ms": 195.0,
+ "active_s": 0.227,
+ "onset_s": 0.033,
+ "env_oct": 1.188,
+ "curve_db": 13.31,
+ "t40_ratio": 1.625,
+ "attack_ratio": 7.0,
+ "active_ratio": 1.039,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 3.96,
+ "phrase_min": 3.61,
+ "margin": 1.85,
+ "wins": 1.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Computer keyboard",
+ 0.264
+ ],
+ [
+ "Typing",
+ 0.17
+ ],
+ [
+ "Typewriter",
+ 0.077
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.264,
+ "floor_db": -40.0,
+ "stereo_corr": 0.603,
+ "darken": 0.7,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -92.1,
+ "first_sample": 0.0,
+ "tail_db": -47.3,
+ "lead_s": 0.004,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/fs-709460_2.wav",
+ "length_s": 0.11,
+ "typ_z": -0.15,
+ "nn_cos": 0.782,
+ "tone_db": 8.13,
+ "tone_ratio": 1.48,
+ "n_outside": 1,
+ "outside": [
+ [
+ "31 Hz",
+ -26.4,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 85.0,
+ "active_s": 0.085,
+ "onset_s": 0.005,
+ "env_oct": 1.394,
+ "curve_db": 12.2,
+ "t40_ratio": 0.708,
+ "attack_ratio": 0.2,
+ "active_ratio": 0.389,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 3.74,
+ "phrase_min": 3.42,
+ "margin": 2.0,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Music",
+ 0.143
+ ],
+ [
+ "Speech",
+ 0.129
+ ],
+ [
+ "Sound effect",
+ 0.045
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.001,
+ "floor_db": -44.2,
+ "stereo_corr": 0.675,
+ "darken": -0.12,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -96.2,
+ "first_sample": 0.0,
+ "tail_db": -50.8,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/fs-665075_1.wav",
+ "length_s": 0.198,
+ "typ_z": 0.97,
+ "nn_cos": 0.875,
+ "tone_db": 5.09,
+ "tone_ratio": 0.93,
+ "n_outside": 1,
+ "outside": [
+ [
+ "5.04 kHz",
+ -10.9,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 150.0,
+ "active_s": 0.166,
+ "onset_s": 0.014,
+ "env_oct": 1.013,
+ "curve_db": 9.28,
+ "t40_ratio": 1.25,
+ "attack_ratio": 0.2,
+ "active_ratio": 0.76,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 4.88,
+ "phrase_min": 4.6,
+ "margin": 3.11,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Computer keyboard",
+ 0.514
+ ],
+ [
+ "Typing",
+ 0.359
+ ],
+ [
+ "Speech",
+ 0.039
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.514,
+ "floor_db": -39.4,
+ "stereo_corr": 1.0,
+ "darken": 1.04,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -57.0,
+ "first_sample": 0.0,
+ "tail_db": -58.0,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/forge/fs-665075_2.wav",
+ "length_s": 0.225,
+ "typ_z": 1.35,
+ "nn_cos": 0.903,
+ "tone_db": 7.38,
+ "tone_ratio": 1.35,
+ "n_outside": 1,
+ "outside": [
+ [
+ "50 Hz",
+ -23.0,
+ "below"
+ ]
+ ],
+ "attack_ms": 8.0,
+ "t40_ms": 130.0,
+ "active_s": 0.135,
+ "onset_s": 0.065,
+ "env_oct": 0.496,
+ "curve_db": 6.73,
+ "t40_ratio": 1.083,
+ "attack_ratio": 1.6,
+ "active_ratio": 0.618,
+ "ref_median": {
+ "attack_ms": 5.0,
+ "t40_ms": 120.0,
+ "active_s": 0.218
+ },
+ "phrase_z": 4.95,
+ "phrase_min": 4.61,
+ "margin": 3.48,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Computer keyboard",
+ 0.687
+ ],
+ [
+ "Typing",
+ 0.517
+ ],
+ [
+ "Sound effect",
+ 0.04
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.687,
+ "floor_db": -42.5,
+ "stereo_corr": 1.0,
+ "darken": 0.45,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -57.0,
+ "first_sample": 0.0,
+ "tail_db": -54.5,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_loo": [
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-378083_1.wav",
+ "length_s": 0.17,
+ "typ_z": 0.15,
+ "nn_cos": 0.822,
+ "tone_db": 4.18,
+ "tone_ratio": 0.73,
+ "n_outside": 1,
+ "outside": [
+ [
+ "1 kHz",
+ -9.3,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 125.0,
+ "active_s": 0.125,
+ "onset_s": 0.035,
+ "env_oct": 1.321,
+ "curve_db": 7.32,
+ "t40_ratio": 1.087,
+ "attack_ratio": 0.143,
+ "active_ratio": 0.488,
+ "ref_median": {
+ "attack_ms": 7.0,
+ "t40_ms": 115.0,
+ "active_s": 0.256
+ },
+ "phrase_z": 3.24,
+ "phrase_min": 2.99,
+ "margin": 0.34,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.151
+ ],
+ [
+ "Computer keyboard",
+ 0.105
+ ],
+ [
+ "Typing",
+ 0.083
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.105,
+ "floor_db": -30.2,
+ "stereo_corr": 0.996,
+ "darken": 0.51,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -76.2,
+ "first_sample": 0.0,
+ "tail_db": -41.6,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-378084_1.wav",
+ "length_s": 0.367,
+ "typ_z": 0.88,
+ "nn_cos": 0.919,
+ "tone_db": 4.16,
+ "tone_ratio": 0.73,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 3.0,
+ "t40_ms": 180.0,
+ "active_s": 0.262,
+ "onset_s": 0.098,
+ "env_oct": 0.801,
+ "curve_db": 17.01,
+ "t40_ratio": 1.565,
+ "attack_ratio": 0.429,
+ "active_ratio": 1.448,
+ "ref_median": {
+ "attack_ms": 7.0,
+ "t40_ms": 115.0,
+ "active_s": 0.181
+ },
+ "phrase_z": 4.55,
+ "phrase_min": 4.29,
+ "margin": 2.2,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.445
+ ],
+ [
+ "Clapping",
+ 0.134
+ ],
+ [
+ "Speech",
+ 0.036
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.01,
+ "floor_db": -39.5,
+ "stereo_corr": 0.997,
+ "darken": 0.87,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -78.0,
+ "first_sample": 0.0,
+ "tail_db": -44.7,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-378085_1.wav",
+ "length_s": 0.16,
+ "typ_z": 0.45,
+ "nn_cos": 0.822,
+ "tone_db": 5.25,
+ "tone_ratio": 0.92,
+ "n_outside": 1,
+ "outside": [
+ [
+ "31 Hz",
+ -12.6,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 75.0,
+ "active_s": 0.126,
+ "onset_s": 0.009,
+ "env_oct": 1.522,
+ "curve_db": 10.62,
+ "t40_ratio": 0.6,
+ "attack_ratio": 0.143,
+ "active_ratio": 0.492,
+ "ref_median": {
+ "attack_ms": 7.0,
+ "t40_ms": 125.0,
+ "active_s": 0.256
+ },
+ "phrase_z": 3.83,
+ "phrase_min": 3.49,
+ "margin": 1.16,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.111
+ ],
+ [
+ "Slap, smack",
+ 0.061
+ ],
+ [
+ "Sound effect",
+ 0.035
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.001,
+ "floor_db": -49.6,
+ "stereo_corr": 0.999,
+ "darken": 0.1,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -79.3,
+ "first_sample": 0.0,
+ "tail_db": -60.5,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-570754_1.wav",
+ "length_s": 0.151,
+ "typ_z": 0.07,
+ "nn_cos": 0.822,
+ "tone_db": 5.26,
+ "tone_ratio": 0.92,
+ "n_outside": 2,
+ "outside": [
+ [
+ "2 kHz",
+ -9.7,
+ "below"
+ ],
+ [
+ "5.04 kHz",
+ 6.4,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 85.0,
+ "active_s": 0.085,
+ "onset_s": 0.045,
+ "env_oct": 1.651,
+ "curve_db": 9.86,
+ "t40_ratio": 0.68,
+ "attack_ratio": 0.143,
+ "active_ratio": 0.332,
+ "ref_median": {
+ "attack_ms": 7.0,
+ "t40_ms": 125.0,
+ "active_s": 0.256
+ },
+ "phrase_z": 3.3,
+ "phrase_min": 2.83,
+ "margin": 1.13,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.176
+ ],
+ [
+ "Coin (dropping)",
+ 0.151
+ ],
+ [
+ "Computer keyboard",
+ 0.049
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.049,
+ "floor_db": -47.8,
+ "stereo_corr": 1.0,
+ "darken": 0.08,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -72.9,
+ "first_sample": 0.0,
+ "tail_db": -56.8,
+ "lead_s": 0.014,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-766639_1.wav",
+ "length_s": 0.104,
+ "typ_z": 0.2,
+ "nn_cos": 0.815,
+ "tone_db": 9.0,
+ "tone_ratio": 1.71,
+ "n_outside": 2,
+ "outside": [
+ [
+ "315 Hz",
+ -26.1,
+ "below"
+ ],
+ [
+ "397 Hz",
+ -25.3,
+ "below"
+ ]
+ ],
+ "attack_ms": 21.0,
+ "t40_ms": 65.0,
+ "active_s": 0.083,
+ "onset_s": 0.002,
+ "env_oct": 1.792,
+ "curve_db": 14.07,
+ "t40_ratio": 0.52,
+ "attack_ratio": 7.0,
+ "active_ratio": 0.324,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 125.0,
+ "active_s": 0.256
+ },
+ "phrase_z": 3.58,
+ "phrase_min": 3.13,
+ "margin": 1.34,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Speech",
+ 0.145
+ ],
+ [
+ "Computer keyboard",
+ 0.077
+ ],
+ [
+ "Finger snapping",
+ 0.069
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.077,
+ "floor_db": -41.2,
+ "stereo_corr": 1.0,
+ "darken": 0.25,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -80.7,
+ "first_sample": 0.0,
+ "tail_db": -50.9,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-766640_1.wav",
+ "length_s": 0.194,
+ "typ_z": 0.11,
+ "nn_cos": 0.815,
+ "tone_db": 10.44,
+ "tone_ratio": 1.98,
+ "n_outside": 8,
+ "outside": [
+ [
+ "198 Hz",
+ -23.8,
+ "below"
+ ],
+ [
+ "157 Hz",
+ -19.8,
+ "below"
+ ],
+ [
+ "125 Hz",
+ -18.1,
+ "below"
+ ]
+ ],
+ "attack_ms": 136.0,
+ "t40_ms": 45.0,
+ "active_s": 0.181,
+ "onset_s": 0.009,
+ "env_oct": 2.492,
+ "curve_db": 19.85,
+ "t40_ratio": 0.36,
+ "attack_ratio": 45.333,
+ "active_ratio": 0.707,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 125.0,
+ "active_s": 0.256
+ },
+ "phrase_z": 3.23,
+ "phrase_min": 2.81,
+ "margin": 0.8,
+ "wins": 1.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.053
+ ],
+ [
+ "Scrape",
+ 0.032
+ ],
+ [
+ "Speech",
+ 0.03
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.0,
+ "floor_db": -35.4,
+ "stereo_corr": 1.0,
+ "darken": -0.67,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -82.5,
+ "first_sample": 0.0,
+ "tail_db": -37.6,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-609534_1.wav",
+ "length_s": 0.348,
+ "typ_z": 0.93,
+ "nn_cos": 0.919,
+ "tone_db": 5.69,
+ "tone_ratio": 1.08,
+ "n_outside": 1,
+ "outside": [
+ [
+ "79 Hz",
+ 12.5,
+ "above"
+ ]
+ ],
+ "attack_ms": 113.0,
+ "t40_ms": 145.0,
+ "active_s": 0.256,
+ "onset_s": 0.074,
+ "env_oct": 2.023,
+ "curve_db": 16.39,
+ "t40_ratio": 1.261,
+ "attack_ratio": 37.667,
+ "active_ratio": 1.414,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 115.0,
+ "active_s": 0.181
+ },
+ "phrase_z": 5.15,
+ "phrase_min": 4.76,
+ "margin": 3.08,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.123
+ ],
+ [
+ "Clapping",
+ 0.088
+ ],
+ [
+ "Ping",
+ 0.041
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.013,
+ "floor_db": -67.5,
+ "stereo_corr": 0.998,
+ "darken": -0.44,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -82.4,
+ "first_sample": 0.0,
+ "tail_db": -51.5,
+ "lead_s": 0.072,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-609534_2.wav",
+ "length_s": 0.442,
+ "typ_z": 0.89,
+ "nn_cos": 0.874,
+ "tone_db": 4.91,
+ "tone_ratio": 0.86,
+ "n_outside": 1,
+ "outside": [
+ [
+ "630 Hz",
+ -11.1,
+ "below"
+ ]
+ ],
+ "attack_ms": 99.0,
+ "t40_ms": 320.0,
+ "active_s": 0.418,
+ "onset_s": 0.007,
+ "env_oct": 2.576,
+ "curve_db": 26.12,
+ "t40_ratio": 2.783,
+ "attack_ratio": 33.0,
+ "active_ratio": 2.309,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 115.0,
+ "active_s": 0.181
+ },
+ "phrase_z": 4.65,
+ "phrase_min": 4.22,
+ "margin": 2.62,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Coin (dropping)",
+ 0.066
+ ],
+ [
+ "Sound effect",
+ 0.059
+ ],
+ [
+ "Single-lens reflex camera",
+ 0.04
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.014,
+ "floor_db": -36.7,
+ "stereo_corr": 0.999,
+ "darken": -2.83,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -83.3,
+ "first_sample": 0.0,
+ "tail_db": -52.1,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-609534_3.wav",
+ "length_s": 0.301,
+ "typ_z": -0.03,
+ "nn_cos": 0.809,
+ "tone_db": 4.86,
+ "tone_ratio": 0.85,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 1.0,
+ "t40_ms": 260.0,
+ "active_s": 0.269,
+ "onset_s": 0.006,
+ "env_oct": 1.519,
+ "curve_db": 20.05,
+ "t40_ratio": 2.261,
+ "attack_ratio": 0.143,
+ "active_ratio": 1.486,
+ "ref_median": {
+ "attack_ms": 7.0,
+ "t40_ms": 115.0,
+ "active_s": 0.181
+ },
+ "phrase_z": 4.84,
+ "phrase_min": 4.36,
+ "margin": 3.09,
+ "wins": 1.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Crack",
+ 0.126
+ ],
+ [
+ "Breaking",
+ 0.059
+ ],
+ [
+ "Sound effect",
+ 0.038
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.001,
+ "floor_db": -39.8,
+ "stereo_corr": 0.999,
+ "darken": 0.16,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -82.6,
+ "first_sample": 0.0,
+ "tail_db": -58.6,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-515810_1.wav",
+ "length_s": 0.319,
+ "typ_z": -12.68,
+ "nn_cos": 0.514,
+ "tone_db": 9.2,
+ "tone_ratio": 1.75,
+ "n_outside": 6,
+ "outside": [
+ [
+ "250 Hz",
+ 21.5,
+ "above"
+ ],
+ [
+ "12.7 kHz",
+ 20.9,
+ "above"
+ ],
+ [
+ "315 Hz",
+ 15.7,
+ "above"
+ ]
+ ],
+ "attack_ms": 7.0,
+ "t40_ms": 105.0,
+ "active_s": 0.288,
+ "onset_s": 0.012,
+ "env_oct": 0.715,
+ "curve_db": 23.77,
+ "t40_ratio": 0.84,
+ "attack_ratio": 2.333,
+ "active_ratio": 1.591,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 125.0,
+ "active_s": 0.181
+ },
+ "phrase_z": 2.32,
+ "phrase_min": 2.06,
+ "margin": 0.55,
+ "wins": 0.5,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.149
+ ],
+ [
+ "Speech",
+ 0.123
+ ],
+ [
+ "Computer keyboard",
+ 0.036
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.036,
+ "floor_db": -40.2,
+ "stereo_corr": 0.994,
+ "darken": -1.14,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -74.7,
+ "first_sample": 0.0,
+ "tail_db": -49.3,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-815614_1.wav",
+ "length_s": 0.269,
+ "typ_z": 0.25,
+ "nn_cos": 0.867,
+ "tone_db": 8.69,
+ "tone_ratio": 1.65,
+ "n_outside": 2,
+ "outside": [
+ [
+ "125 Hz",
+ 18.5,
+ "above"
+ ],
+ [
+ "99 Hz",
+ 12.3,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 260.0,
+ "active_s": 0.256,
+ "onset_s": 0.009,
+ "env_oct": 1.495,
+ "curve_db": 17.78,
+ "t40_ratio": 2.261,
+ "attack_ratio": 0.143,
+ "active_ratio": 1.414,
+ "ref_median": {
+ "attack_ms": 7.0,
+ "t40_ms": 115.0,
+ "active_s": 0.181
+ },
+ "phrase_z": 3.47,
+ "phrase_min": 3.2,
+ "margin": 1.72,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Tap",
+ 0.119
+ ],
+ [
+ "Music",
+ 0.116
+ ],
+ [
+ "Sound effect",
+ 0.088
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.016,
+ "floor_db": -27.3,
+ "stereo_corr": 1.0,
+ "darken": -0.09,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -70.5,
+ "first_sample": 0.0,
+ "tail_db": -31.0,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-815614_2.wav",
+ "length_s": 0.181,
+ "typ_z": 0.52,
+ "nn_cos": 0.867,
+ "tone_db": 8.02,
+ "tone_ratio": 1.52,
+ "n_outside": 1,
+ "outside": [
+ [
+ "794 Hz",
+ -6.8,
+ "below"
+ ]
+ ],
+ "attack_ms": 7.0,
+ "t40_ms": 115.0,
+ "active_s": 0.135,
+ "onset_s": 0.04,
+ "env_oct": 0.755,
+ "curve_db": 6.9,
+ "t40_ratio": 0.92,
+ "attack_ratio": 2.333,
+ "active_ratio": 0.527,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 125.0,
+ "active_s": 0.256
+ },
+ "phrase_z": 3.93,
+ "phrase_min": 3.57,
+ "margin": 2.23,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Computer keyboard",
+ 0.094
+ ],
+ [
+ "Speech",
+ 0.065
+ ],
+ [
+ "Tap",
+ 0.057
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.094,
+ "floor_db": -37.6,
+ "stereo_corr": 1.0,
+ "darken": 1.32,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -74.7,
+ "first_sample": 0.0,
+ "tail_db": -41.8,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_rev": [
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-378083_1.wav",
+ "length_s": 0.17,
+ "typ_z": -1.72,
+ "nn_cos": 0.748,
+ "tone_db": 6.31,
+ "tone_ratio": 0.74,
+ "n_outside": 1,
+ "outside": [
+ [
+ "10.1 kHz",
+ -9.8,
+ "below"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 125.0,
+ "active_s": 0.125,
+ "onset_s": 0.035,
+ "env_oct": 0.768,
+ "curve_db": 12.15,
+ "t40_ratio": 0.806,
+ "attack_ratio": 0.333,
+ "active_ratio": 0.753,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 155.0,
+ "active_s": 0.166
+ },
+ "phrase_z": 3.24,
+ "phrase_min": 2.99,
+ "margin": 0.34,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.151
+ ],
+ [
+ "Computer keyboard",
+ 0.105
+ ],
+ [
+ "Typing",
+ 0.083
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.105,
+ "floor_db": -30.2,
+ "stereo_corr": 0.996,
+ "darken": 0.51,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -76.2,
+ "first_sample": 0.0,
+ "tail_db": -41.6,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-378084_1.wav",
+ "length_s": 0.367,
+ "typ_z": 0.49,
+ "nn_cos": 0.892,
+ "tone_db": 8.06,
+ "tone_ratio": 0.95,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 3.0,
+ "t40_ms": 180.0,
+ "active_s": 0.262,
+ "onset_s": 0.098,
+ "env_oct": 0.291,
+ "curve_db": 16.7,
+ "t40_ratio": 1.161,
+ "attack_ratio": 1.0,
+ "active_ratio": 1.578,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 155.0,
+ "active_s": 0.166
+ },
+ "phrase_z": 4.55,
+ "phrase_min": 4.29,
+ "margin": 2.2,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.445
+ ],
+ [
+ "Clapping",
+ 0.134
+ ],
+ [
+ "Speech",
+ 0.036
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.01,
+ "floor_db": -39.5,
+ "stereo_corr": 0.997,
+ "darken": 0.87,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -78.0,
+ "first_sample": 0.0,
+ "tail_db": -44.7,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-378085_1.wav",
+ "length_s": 0.16,
+ "typ_z": -1.31,
+ "nn_cos": 0.789,
+ "tone_db": 5.85,
+ "tone_ratio": 0.69,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 1.0,
+ "t40_ms": 75.0,
+ "active_s": 0.126,
+ "onset_s": 0.009,
+ "env_oct": 1.01,
+ "curve_db": 10.48,
+ "t40_ratio": 0.484,
+ "attack_ratio": 0.333,
+ "active_ratio": 0.759,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 155.0,
+ "active_s": 0.166
+ },
+ "phrase_z": 3.83,
+ "phrase_min": 3.49,
+ "margin": 1.16,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.111
+ ],
+ [
+ "Slap, smack",
+ 0.061
+ ],
+ [
+ "Sound effect",
+ 0.035
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.001,
+ "floor_db": -49.6,
+ "stereo_corr": 0.999,
+ "darken": 0.1,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -79.3,
+ "first_sample": 0.0,
+ "tail_db": -60.5,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-570754_1.wav",
+ "length_s": 0.151,
+ "typ_z": -1.82,
+ "nn_cos": 0.75,
+ "tone_db": 8.34,
+ "tone_ratio": 0.98,
+ "n_outside": 1,
+ "outside": [
+ [
+ "31 Hz",
+ 25.9,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 85.0,
+ "active_s": 0.085,
+ "onset_s": 0.045,
+ "env_oct": 1.139,
+ "curve_db": 14.5,
+ "t40_ratio": 0.548,
+ "attack_ratio": 0.333,
+ "active_ratio": 0.512,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 155.0,
+ "active_s": 0.166
+ },
+ "phrase_z": 3.3,
+ "phrase_min": 2.83,
+ "margin": 1.13,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.176
+ ],
+ [
+ "Coin (dropping)",
+ 0.151
+ ],
+ [
+ "Computer keyboard",
+ 0.049
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.049,
+ "floor_db": -47.8,
+ "stereo_corr": 1.0,
+ "darken": 0.08,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -72.9,
+ "first_sample": 0.0,
+ "tail_db": -56.8,
+ "lead_s": 0.014,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-766639_1.wav",
+ "length_s": 0.104,
+ "typ_z": -1.14,
+ "nn_cos": 0.789,
+ "tone_db": 8.4,
+ "tone_ratio": 0.99,
+ "n_outside": 3,
+ "outside": [
+ [
+ "315 Hz",
+ -21.5,
+ "below"
+ ],
+ [
+ "397 Hz",
+ -21.1,
+ "below"
+ ],
+ [
+ "630 Hz",
+ 12.9,
+ "above"
+ ]
+ ],
+ "attack_ms": 21.0,
+ "t40_ms": 65.0,
+ "active_s": 0.083,
+ "onset_s": 0.002,
+ "env_oct": 1.687,
+ "curve_db": 17.63,
+ "t40_ratio": 0.419,
+ "attack_ratio": 7.0,
+ "active_ratio": 0.5,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 155.0,
+ "active_s": 0.166
+ },
+ "phrase_z": 3.58,
+ "phrase_min": 3.13,
+ "margin": 1.34,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Speech",
+ 0.145
+ ],
+ [
+ "Computer keyboard",
+ 0.077
+ ],
+ [
+ "Finger snapping",
+ 0.069
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.077,
+ "floor_db": -41.2,
+ "stereo_corr": 1.0,
+ "darken": 0.25,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -80.7,
+ "first_sample": 0.0,
+ "tail_db": -50.9,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-766640_1.wav",
+ "length_s": 0.194,
+ "typ_z": -1.32,
+ "nn_cos": 0.771,
+ "tone_db": 10.12,
+ "tone_ratio": 1.19,
+ "n_outside": 2,
+ "outside": [
+ [
+ "198 Hz",
+ -18.9,
+ "below"
+ ],
+ [
+ "1 kHz",
+ 9.6,
+ "above"
+ ]
+ ],
+ "attack_ms": 136.0,
+ "t40_ms": 45.0,
+ "active_s": 0.181,
+ "onset_s": 0.009,
+ "env_oct": 2.471,
+ "curve_db": 24.3,
+ "t40_ratio": 0.29,
+ "attack_ratio": 45.333,
+ "active_ratio": 1.09,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 155.0,
+ "active_s": 0.166
+ },
+ "phrase_z": 3.23,
+ "phrase_min": 2.81,
+ "margin": 0.8,
+ "wins": 1.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.053
+ ],
+ [
+ "Scrape",
+ 0.032
+ ],
+ [
+ "Speech",
+ 0.03
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.0,
+ "floor_db": -35.4,
+ "stereo_corr": 1.0,
+ "darken": -0.67,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -82.5,
+ "first_sample": 0.0,
+ "tail_db": -37.6,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-609534_1.wav",
+ "length_s": 0.348,
+ "typ_z": 0.99,
+ "nn_cos": 0.927,
+ "tone_db": 5.06,
+ "tone_ratio": 0.59,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 113.0,
+ "t40_ms": 145.0,
+ "active_s": 0.256,
+ "onset_s": 0.074,
+ "env_oct": 1.985,
+ "curve_db": 19.66,
+ "t40_ratio": 0.935,
+ "attack_ratio": 37.667,
+ "active_ratio": 1.542,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 155.0,
+ "active_s": 0.166
+ },
+ "phrase_z": 5.15,
+ "phrase_min": 4.76,
+ "margin": 3.08,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Finger snapping",
+ 0.123
+ ],
+ [
+ "Clapping",
+ 0.088
+ ],
+ [
+ "Ping",
+ 0.041
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.013,
+ "floor_db": -67.5,
+ "stereo_corr": 0.998,
+ "darken": -0.44,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -82.4,
+ "first_sample": 0.0,
+ "tail_db": -51.5,
+ "lead_s": 0.072,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-609534_2.wav",
+ "length_s": 0.442,
+ "typ_z": 0.51,
+ "nn_cos": 0.903,
+ "tone_db": 6.7,
+ "tone_ratio": 0.79,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 99.0,
+ "t40_ms": 320.0,
+ "active_s": 0.418,
+ "onset_s": 0.007,
+ "env_oct": 2.474,
+ "curve_db": 26.86,
+ "t40_ratio": 2.065,
+ "attack_ratio": 33.0,
+ "active_ratio": 2.518,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 155.0,
+ "active_s": 0.166
+ },
+ "phrase_z": 4.65,
+ "phrase_min": 4.22,
+ "margin": 2.62,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Coin (dropping)",
+ 0.066
+ ],
+ [
+ "Sound effect",
+ 0.059
+ ],
+ [
+ "Single-lens reflex camera",
+ 0.04
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.014,
+ "floor_db": -36.7,
+ "stereo_corr": 0.999,
+ "darken": -2.83,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -83.3,
+ "first_sample": 0.0,
+ "tail_db": -52.1,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-609534_3.wav",
+ "length_s": 0.301,
+ "typ_z": -0.83,
+ "nn_cos": 0.823,
+ "tone_db": 5.61,
+ "tone_ratio": 0.66,
+ "n_outside": 1,
+ "outside": [
+ [
+ "1 kHz",
+ 11.3,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 260.0,
+ "active_s": 0.269,
+ "onset_s": 0.006,
+ "env_oct": 1.009,
+ "curve_db": 15.71,
+ "t40_ratio": 1.677,
+ "attack_ratio": 0.333,
+ "active_ratio": 1.62,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 155.0,
+ "active_s": 0.166
+ },
+ "phrase_z": 4.84,
+ "phrase_min": 4.36,
+ "margin": 3.09,
+ "wins": 1.0,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Crack",
+ 0.126
+ ],
+ [
+ "Breaking",
+ 0.059
+ ],
+ [
+ "Sound effect",
+ 0.038
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.001,
+ "floor_db": -39.8,
+ "stereo_corr": 0.999,
+ "darken": 0.16,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -82.6,
+ "first_sample": 0.0,
+ "tail_db": -58.6,
+ "lead_s": 0.002,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-515810_1.wav",
+ "length_s": 0.319,
+ "typ_z": -4.1,
+ "nn_cos": 0.583,
+ "tone_db": 11.19,
+ "tone_ratio": 1.32,
+ "n_outside": 5,
+ "outside": [
+ [
+ "250 Hz",
+ 27.4,
+ "above"
+ ],
+ [
+ "397 Hz",
+ 21.6,
+ "above"
+ ],
+ [
+ "315 Hz",
+ 20.3,
+ "above"
+ ]
+ ],
+ "attack_ms": 7.0,
+ "t40_ms": 105.0,
+ "active_s": 0.288,
+ "onset_s": 0.012,
+ "env_oct": 0.86,
+ "curve_db": 22.11,
+ "t40_ratio": 0.677,
+ "attack_ratio": 2.333,
+ "active_ratio": 1.735,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 155.0,
+ "active_s": 0.166
+ },
+ "phrase_z": 2.32,
+ "phrase_min": 2.06,
+ "margin": 0.55,
+ "wins": 0.5,
+ "best_confusion": "a stapler",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.149
+ ],
+ [
+ "Speech",
+ 0.123
+ ],
+ [
+ "Computer keyboard",
+ 0.036
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.036,
+ "floor_db": -40.2,
+ "stereo_corr": 0.994,
+ "darken": -1.14,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -74.7,
+ "first_sample": 0.0,
+ "tail_db": -49.3,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-815614_1.wav",
+ "length_s": 0.269,
+ "typ_z": -1.45,
+ "nn_cos": 0.764,
+ "tone_db": 11.51,
+ "tone_ratio": 1.35,
+ "n_outside": 5,
+ "outside": [
+ [
+ "198 Hz",
+ 22.6,
+ "above"
+ ],
+ [
+ "397 Hz",
+ 19.8,
+ "above"
+ ],
+ [
+ "157 Hz",
+ 19.7,
+ "above"
+ ]
+ ],
+ "attack_ms": 1.0,
+ "t40_ms": 260.0,
+ "active_s": 0.256,
+ "onset_s": 0.009,
+ "env_oct": 0.985,
+ "curve_db": 20.98,
+ "t40_ratio": 1.677,
+ "attack_ratio": 0.333,
+ "active_ratio": 1.542,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 155.0,
+ "active_s": 0.166
+ },
+ "phrase_z": 3.47,
+ "phrase_min": 3.2,
+ "margin": 1.72,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Tap",
+ 0.119
+ ],
+ [
+ "Music",
+ 0.116
+ ],
+ [
+ "Sound effect",
+ 0.088
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.016,
+ "floor_db": -27.3,
+ "stereo_corr": 1.0,
+ "darken": -0.09,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -70.5,
+ "first_sample": 0.0,
+ "tail_db": -31.0,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/typing_key/bench/fs-815614_2.wav",
+ "length_s": 0.181,
+ "typ_z": -0.61,
+ "nn_cos": 0.841,
+ "tone_db": 10.73,
+ "tone_ratio": 1.26,
+ "n_outside": 3,
+ "outside": [
+ [
+ "31 Hz",
+ 27.0,
+ "above"
+ ],
+ [
+ "198 Hz",
+ 24.3,
+ "above"
+ ],
+ [
+ "157 Hz",
+ 20.3,
+ "above"
+ ]
+ ],
+ "attack_ms": 7.0,
+ "t40_ms": 115.0,
+ "active_s": 0.135,
+ "onset_s": 0.04,
+ "env_oct": 0.65,
+ "curve_db": 13.11,
+ "t40_ratio": 0.742,
+ "attack_ratio": 2.333,
+ "active_ratio": 0.813,
+ "ref_median": {
+ "attack_ms": 3.0,
+ "t40_ms": 155.0,
+ "active_s": 0.166
+ },
+ "phrase_z": 3.93,
+ "phrase_min": 3.57,
+ "margin": 2.23,
+ "wins": 1.0,
+ "best_confusion": "a ballpoint pen click",
+ "as_top": [
+ [
+ "Computer keyboard",
+ 0.094
+ ],
+ [
+ "Speech",
+ 0.065
+ ],
+ [
+ "Tap",
+ 0.057
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.094,
+ "floor_db": -37.6,
+ "stereo_corr": 1.0,
+ "darken": 1.32,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -74.7,
+ "first_sample": 0.0,
+ "tail_db": -41.8,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity_bench": {
+ "n": 12,
+ "cos_mean": 0.673,
+ "cos_max": 0.919,
+ "tone_min_db": 4.64,
+ "tone_mean_db": 9.33,
+ "near_duplicates": []
+ },
+ "diversity_forge": {
+ "n": 15,
+ "cos_mean": 0.721,
+ "cos_max": 0.978,
+ "tone_min_db": 3.07,
+ "tone_mean_db": 12.39,
+ "near_duplicates": []
+ },
+ "inside": {
+ "per_sound": [
+ 7,
+ 7,
+ 7
+ ],
+ "n_checks": 7,
+ "limits": {
+ "typ_z": -0.15,
+ "tone_db": 10.17,
+ "env_oct": 1.521,
+ "curve_db": 19.148,
+ "phrase_z": 3.358,
+ "margin": 1.176,
+ "as_p": 0.008
+ },
+ "real_per_sound": [
+ 7,
+ 5,
+ 7,
+ 7,
+ 6,
+ 7,
+ 4,
+ 5,
+ 5,
+ 6,
+ 2,
+ 7,
+ 6,
+ 7,
+ 7,
+ 5,
+ 7,
+ 4,
+ 4,
+ 6,
+ 1,
+ 6,
+ 5,
+ 5,
+ 3,
+ 7,
+ 7
+ ]
+ },
+ "summary": {
+ "made": {
+ "n": 3,
+ "typ_z": 1.2,
+ "tone_db": 8.2,
+ "env_oct": 1.07,
+ "curve_db": 14.32,
+ "phrase_z": 5.56,
+ "margin": 3.57,
+ "as_p": 0.07,
+ "tone_ratio": 1.5,
+ "t40_ratio": 1.67,
+ "wins": 1.0,
+ "floor_db": -54.7,
+ "stereo_corr": 0.94,
+ "darken": 0.4,
+ "as_names": 0.33,
+ "artifacts": 0,
+ "var_cos": 0.965
+ },
+ "real_cross": {
+ "n": 15,
+ "typ_z": 0.26,
+ "tone_db": 8.88,
+ "env_oct": 1.01,
+ "curve_db": 12.2,
+ "phrase_z": 4.75,
+ "margin": 2.83,
+ "as_p": 0.08,
+ "tone_ratio": 1.62,
+ "t40_ratio": 1.29,
+ "wins": 1.0,
+ "floor_db": -43.7,
+ "stereo_corr": 0.76,
+ "darken": 0.04,
+ "as_names": 0.53,
+ "artifacts": 0,
+ "var_cos": 0.721
+ },
+ "real_loo": {
+ "n": 12,
+ "typ_z": 0.23,
+ "tone_db": 5.47,
+ "env_oct": 1.52,
+ "curve_db": 16.7,
+ "phrase_z": 3.71,
+ "margin": 1.53,
+ "as_p": 0.01,
+ "tone_ratio": 1.0,
+ "t40_ratio": 1.0,
+ "wins": 1.0,
+ "floor_db": -39.65,
+ "stereo_corr": 1.0,
+ "darken": 0.09,
+ "as_names": 0.42,
+ "artifacts": 0,
+ "var_cos": 0.673
+ }
+ }
+ },
+ "whoosh_fast": {
+ "n_bench": 5,
+ "n_forge": 5,
+ "note": "",
+ "bench_refs": [
+ "E:/earshot/out/refs/whoosh_fast/bench/fs-352719_1.wav",
+ "E:/earshot/out/refs/whoosh_fast/bench/fs-216675_1.wav",
+ "E:/earshot/out/refs/whoosh_fast/bench/fs-463763_1.wav",
+ "E:/earshot/out/refs/whoosh_fast/bench/fs-463763_2.wav",
+ "E:/earshot/out/refs/whoosh_fast/bench/fs-463763_3.wav"
+ ],
+ "made": [
+ {
+ "file": "demo/forge-test/whoosh/final/whoosh_01.wav",
+ "length_s": 0.313,
+ "typ_z": 2.01,
+ "nn_cos": 0.912,
+ "tone_db": 8.58,
+ "tone_ratio": 1.87,
+ "n_outside": 8,
+ "outside": [
+ [
+ "4 kHz",
+ 11.4,
+ "above"
+ ],
+ [
+ "5.04 kHz",
+ 10.4,
+ "above"
+ ],
+ [
+ "6.35 kHz",
+ 10.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 51.0,
+ "t40_ms": 125.0,
+ "active_s": 0.246,
+ "onset_s": 0.034,
+ "env_oct": 0.888,
+ "curve_db": 15.93,
+ "t40_ratio": 0.472,
+ "attack_ratio": 0.472,
+ "active_ratio": 0.709,
+ "ref_median": {
+ "attack_ms": 108.0,
+ "t40_ms": 265.0,
+ "active_s": 0.347
+ },
+ "phrase_z": 4.47,
+ "phrase_min": 4.31,
+ "margin": 3.31,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.208
+ ],
+ [
+ "Whack, thwack",
+ 0.082
+ ],
+ [
+ "Bang",
+ 0.076
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.023,
+ "floor_db": -43.9,
+ "stereo_corr": 0.9,
+ "darken": -0.19,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -94.2,
+ "first_sample": 0.0,
+ "tail_db": -70.3,
+ "lead_s": 0.003,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "demo/forge-test/whoosh/final/whoosh_02.wav",
+ "length_s": 0.394,
+ "typ_z": 2.05,
+ "nn_cos": 0.929,
+ "tone_db": 6.88,
+ "tone_ratio": 1.5,
+ "n_outside": 2,
+ "outside": [
+ [
+ "250 Hz",
+ 15.8,
+ "above"
+ ],
+ [
+ "79 Hz",
+ 5.7,
+ "above"
+ ]
+ ],
+ "attack_ms": 35.0,
+ "t40_ms": 160.0,
+ "active_s": 0.218,
+ "onset_s": 0.087,
+ "env_oct": 1.008,
+ "curve_db": 13.85,
+ "t40_ratio": 0.604,
+ "attack_ratio": 0.324,
+ "active_ratio": 0.628,
+ "ref_median": {
+ "attack_ms": 108.0,
+ "t40_ms": 265.0,
+ "active_s": 0.347
+ },
+ "phrase_z": 4.42,
+ "phrase_min": 4.27,
+ "margin": 3.52,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Whack, thwack",
+ 0.066
+ ],
+ [
+ "Whoosh, swoosh, swish",
+ 0.06
+ ],
+ [
+ "Sound effect",
+ 0.053
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.06,
+ "floor_db": -60.1,
+ "stereo_corr": 0.858,
+ "darken": -0.71,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -81.5,
+ "first_sample": 0.0,
+ "tail_db": -73.7,
+ "lead_s": 0.049,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "demo/forge-test/whoosh/final/whoosh_03.wav",
+ "length_s": 0.247,
+ "typ_z": 2.24,
+ "nn_cos": 0.912,
+ "tone_db": 8.87,
+ "tone_ratio": 1.94,
+ "n_outside": 8,
+ "outside": [
+ [
+ "794 Hz",
+ -15.7,
+ "below"
+ ],
+ [
+ "315 Hz",
+ 14.1,
+ "above"
+ ],
+ [
+ "157 Hz",
+ 10.6,
+ "above"
+ ]
+ ],
+ "attack_ms": 62.0,
+ "t40_ms": 135.0,
+ "active_s": 0.194,
+ "onset_s": 0.021,
+ "env_oct": 0.871,
+ "curve_db": 20.07,
+ "t40_ratio": 0.509,
+ "attack_ratio": 0.574,
+ "active_ratio": 0.559,
+ "ref_median": {
+ "attack_ms": 108.0,
+ "t40_ms": 265.0,
+ "active_s": 0.347
+ },
+ "phrase_z": 4.46,
+ "phrase_min": 4.24,
+ "margin": 3.2,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.155
+ ],
+ [
+ "Music",
+ 0.073
+ ],
+ [
+ "Whack, thwack",
+ 0.051
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.009,
+ "floor_db": -43.6,
+ "stereo_corr": 0.956,
+ "darken": -0.32,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -71.6,
+ "first_sample": 0.0,
+ "tail_db": -62.2,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity": {
+ "n": 3,
+ "cos_mean": 0.957,
+ "cos_max": 0.973,
+ "tone_min_db": 4.11,
+ "tone_mean_db": 4.69,
+ "near_duplicates": []
+ },
+ "real_cross": [
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/forge/kit-whoosh_fast_01.wav",
+ "length_s": 0.315,
+ "typ_z": -2.82,
+ "nn_cos": 0.805,
+ "tone_db": 10.5,
+ "tone_ratio": 2.29,
+ "n_outside": 9,
+ "outside": [
+ [
+ "10.1 kHz",
+ 14.5,
+ "above"
+ ],
+ [
+ "8 kHz",
+ 14.2,
+ "above"
+ ],
+ [
+ "12.7 kHz",
+ 14.2,
+ "above"
+ ]
+ ],
+ "attack_ms": 31.0,
+ "t40_ms": 105.0,
+ "active_s": 0.139,
+ "onset_s": 0.031,
+ "env_oct": 1.485,
+ "curve_db": 22.56,
+ "t40_ratio": 0.396,
+ "attack_ratio": 0.287,
+ "active_ratio": 0.401,
+ "ref_median": {
+ "attack_ms": 108.0,
+ "t40_ms": 265.0,
+ "active_s": 0.347
+ },
+ "phrase_z": 3.54,
+ "phrase_min": 3.04,
+ "margin": 2.29,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.27
+ ],
+ [
+ "Bang",
+ 0.108
+ ],
+ [
+ "Whack, thwack",
+ 0.09
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.007,
+ "floor_db": -65.9,
+ "stereo_corr": 0.814,
+ "darken": -0.16,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -49.0,
+ "first_sample": 0.0,
+ "tail_db": -101.0,
+ "lead_s": 0.013,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/forge/kit-whoosh_fast_02.wav",
+ "length_s": 0.5,
+ "typ_z": 1.26,
+ "nn_cos": 0.926,
+ "tone_db": 7.4,
+ "tone_ratio": 1.62,
+ "n_outside": 1,
+ "outside": [
+ [
+ "10.1 kHz",
+ 11.2,
+ "above"
+ ]
+ ],
+ "attack_ms": 72.0,
+ "t40_ms": 180.0,
+ "active_s": 0.25,
+ "onset_s": 0.09,
+ "env_oct": 0.539,
+ "curve_db": 9.22,
+ "t40_ratio": 0.679,
+ "attack_ratio": 0.667,
+ "active_ratio": 0.72,
+ "ref_median": {
+ "attack_ms": 108.0,
+ "t40_ms": 265.0,
+ "active_s": 0.347
+ },
+ "phrase_z": 4.27,
+ "phrase_min": 4.04,
+ "margin": 3.39,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Zipper (clothing)",
+ 0.343
+ ],
+ [
+ "Whoosh, swoosh, swish",
+ 0.238
+ ],
+ [
+ "Sound effect",
+ 0.118
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.238,
+ "floor_db": -64.7,
+ "stereo_corr": -0.162,
+ "darken": -0.24,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -112.1,
+ "first_sample": 0.0,
+ "tail_db": -81.3,
+ "lead_s": 0.029,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/forge/kit-whoosh_fast_03.wav",
+ "length_s": 0.39,
+ "typ_z": 1.15,
+ "nn_cos": 0.908,
+ "tone_db": 10.31,
+ "tone_ratio": 2.25,
+ "n_outside": 12,
+ "outside": [
+ [
+ "794 Hz",
+ -20.5,
+ "below"
+ ],
+ [
+ "198 Hz",
+ 18.5,
+ "above"
+ ],
+ [
+ "157 Hz",
+ 16.8,
+ "above"
+ ]
+ ],
+ "attack_ms": 51.0,
+ "t40_ms": 90.0,
+ "active_s": 0.141,
+ "onset_s": 0.094,
+ "env_oct": 1.313,
+ "curve_db": 21.2,
+ "t40_ratio": 0.34,
+ "attack_ratio": 0.472,
+ "active_ratio": 0.406,
+ "ref_median": {
+ "attack_ms": 108.0,
+ "t40_ms": 265.0,
+ "active_s": 0.347
+ },
+ "phrase_z": 4.6,
+ "phrase_min": 4.27,
+ "margin": 3.4,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Zipper (clothing)",
+ 0.211
+ ],
+ [
+ "Whoosh, swoosh, swish",
+ 0.079
+ ],
+ [
+ "Sound effect",
+ 0.048
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.079,
+ "floor_db": -65.4,
+ "stereo_corr": 0.927,
+ "darken": 0.88,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -68.5,
+ "first_sample": 0.0,
+ "tail_db": -85.1,
+ "lead_s": 0.005,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/forge/kit-whoosh_fast_04.wav",
+ "length_s": 0.395,
+ "typ_z": -2.65,
+ "nn_cos": 0.762,
+ "tone_db": 12.8,
+ "tone_ratio": 2.8,
+ "n_outside": 12,
+ "outside": [
+ [
+ "31 Hz",
+ -24.9,
+ "below"
+ ],
+ [
+ "12.7 kHz",
+ 21.4,
+ "above"
+ ],
+ [
+ "10.1 kHz",
+ 21.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 24.0,
+ "t40_ms": 160.0,
+ "active_s": 0.203,
+ "onset_s": 0.037,
+ "env_oct": 1.224,
+ "curve_db": 18.13,
+ "t40_ratio": 0.604,
+ "attack_ratio": 0.222,
+ "active_ratio": 0.585,
+ "ref_median": {
+ "attack_ms": 108.0,
+ "t40_ms": 265.0,
+ "active_s": 0.347
+ },
+ "phrase_z": 3.72,
+ "phrase_min": 3.41,
+ "margin": 2.75,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.172
+ ],
+ [
+ "Whack, thwack",
+ 0.085
+ ],
+ [
+ "Bang",
+ 0.075
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.056,
+ "floor_db": -67.3,
+ "stereo_corr": 0.919,
+ "darken": -0.54,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -76.8,
+ "first_sample": 0.0,
+ "tail_db": -86.8,
+ "lead_s": 0.004,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/forge/fs-719636_1.wav",
+ "length_s": 0.275,
+ "typ_z": 0.57,
+ "nn_cos": 0.895,
+ "tone_db": 10.03,
+ "tone_ratio": 2.19,
+ "n_outside": 2,
+ "outside": [
+ [
+ "250 Hz",
+ 16.3,
+ "above"
+ ],
+ [
+ "500 Hz",
+ -7.6,
+ "below"
+ ]
+ ],
+ "attack_ms": 22.0,
+ "t40_ms": 210.0,
+ "active_s": 0.228,
+ "onset_s": 0.017,
+ "env_oct": 1.079,
+ "curve_db": 20.2,
+ "t40_ratio": 0.792,
+ "attack_ratio": 0.204,
+ "active_ratio": 0.657,
+ "ref_median": {
+ "attack_ms": 108.0,
+ "t40_ms": 265.0,
+ "active_s": 0.347
+ },
+ "phrase_z": 4.14,
+ "phrase_min": 3.95,
+ "margin": 2.78,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Sound effect",
+ 0.163
+ ],
+ [
+ "Bang",
+ 0.082
+ ],
+ [
+ "Whack, thwack",
+ 0.066
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.006,
+ "floor_db": -40.3,
+ "stereo_corr": 0.605,
+ "darken": 0.64,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -44.4,
+ "first_sample": 0.0,
+ "tail_db": -69.6,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_loo": [
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/bench/fs-352719_1.wav",
+ "length_s": 0.435,
+ "typ_z": -0.1,
+ "nn_cos": 0.908,
+ "tone_db": 9.54,
+ "tone_ratio": 2.14,
+ "n_outside": 5,
+ "outside": [
+ [
+ "31 Hz",
+ 23.5,
+ "above"
+ ],
+ [
+ "1 kHz",
+ -23.3,
+ "below"
+ ],
+ [
+ "39 Hz",
+ 15.1,
+ "above"
+ ]
+ ],
+ "attack_ms": 41.0,
+ "t40_ms": 220.0,
+ "active_s": 0.325,
+ "onset_s": 0.065,
+ "env_oct": 0.601,
+ "curve_db": 8.9,
+ "t40_ratio": 0.822,
+ "attack_ratio": 0.373,
+ "active_ratio": 0.935,
+ "ref_median": {
+ "attack_ms": 110.0,
+ "t40_ms": 267.5,
+ "active_s": 0.347
+ },
+ "phrase_z": 4.44,
+ "phrase_min": 4.24,
+ "margin": 3.67,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.102
+ ],
+ [
+ "Sound effect",
+ 0.08
+ ],
+ [
+ "Zipper (clothing)",
+ 0.035
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.102,
+ "floor_db": -44.6,
+ "stereo_corr": 1.0,
+ "darken": -0.45,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -54.7,
+ "first_sample": 0.0,
+ "tail_db": -64.7,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/bench/fs-216675_1.wav",
+ "length_s": 0.23,
+ "typ_z": -0.43,
+ "nn_cos": 0.908,
+ "tone_db": 10.24,
+ "tone_ratio": 2.29,
+ "n_outside": 11,
+ "outside": [
+ [
+ "50 Hz",
+ -23.9,
+ "below"
+ ],
+ [
+ "39 Hz",
+ -17.4,
+ "below"
+ ],
+ [
+ "397 Hz",
+ 13.6,
+ "above"
+ ]
+ ],
+ "attack_ms": 33.0,
+ "t40_ms": 110.0,
+ "active_s": 0.177,
+ "onset_s": 0.023,
+ "env_oct": 1.331,
+ "curve_db": 23.4,
+ "t40_ratio": 0.411,
+ "attack_ratio": 0.3,
+ "active_ratio": 0.509,
+ "ref_median": {
+ "attack_ms": 110.0,
+ "t40_ms": 267.5,
+ "active_s": 0.347
+ },
+ "phrase_z": 4.34,
+ "phrase_min": 3.89,
+ "margin": 3.35,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Music",
+ 0.209
+ ],
+ [
+ "Sound effect",
+ 0.206
+ ],
+ [
+ "Bang",
+ 0.138
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.005,
+ "floor_db": -46.7,
+ "stereo_corr": 1.0,
+ "darken": -0.77,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -98.6,
+ "first_sample": 0.0,
+ "tail_db": -63.8,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/bench/fs-463763_1.wav",
+ "length_s": 0.435,
+ "typ_z": 1.52,
+ "nn_cos": 0.915,
+ "tone_db": 4.58,
+ "tone_ratio": 0.66,
+ "n_outside": 1,
+ "outside": [
+ [
+ "1.26 kHz",
+ 11.5,
+ "above"
+ ]
+ ],
+ "attack_ms": 117.0,
+ "t40_ms": 295.0,
+ "active_s": 0.406,
+ "onset_s": 0.009,
+ "env_oct": 0.402,
+ "curve_db": 8.21,
+ "t40_ratio": 1.216,
+ "attack_ratio": 1.57,
+ "active_ratio": 1.208,
+ "ref_median": {
+ "attack_ms": 74.5,
+ "t40_ms": 242.5,
+ "active_s": 0.336
+ },
+ "phrase_z": 3.92,
+ "phrase_min": 3.86,
+ "margin": 2.32,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.091
+ ],
+ [
+ "Whip",
+ 0.073
+ ],
+ [
+ "Sound effect",
+ 0.063
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.091,
+ "floor_db": -32.2,
+ "stereo_corr": 0.995,
+ "darken": -0.03,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -63.4,
+ "first_sample": 0.0,
+ "tail_db": -49.2,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/bench/fs-463763_2.wav",
+ "length_s": 0.398,
+ "typ_z": 6.7,
+ "nn_cos": 0.951,
+ "tone_db": 4.35,
+ "tone_ratio": 0.62,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 108.0,
+ "t40_ms": 270.0,
+ "active_s": 0.347,
+ "onset_s": 0.033,
+ "env_oct": 0.232,
+ "curve_db": 5.81,
+ "t40_ratio": 1.113,
+ "attack_ratio": 1.412,
+ "active_ratio": 1.031,
+ "ref_median": {
+ "attack_ms": 76.5,
+ "t40_ms": 242.5,
+ "active_s": 0.337
+ },
+ "phrase_z": 4.14,
+ "phrase_min": 4.01,
+ "margin": 2.62,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.138
+ ],
+ [
+ "Sound effect",
+ 0.063
+ ],
+ [
+ "Music",
+ 0.033
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.138,
+ "floor_db": -28.7,
+ "stereo_corr": 0.991,
+ "darken": -0.18,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -51.9,
+ "first_sample": 0.0,
+ "tail_db": -49.3,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/bench/fs-463763_3.wav",
+ "length_s": 0.392,
+ "typ_z": 3.93,
+ "nn_cos": 0.951,
+ "tone_db": 4.26,
+ "tone_ratio": 0.6,
+ "n_outside": 0,
+ "outside": [],
+ "attack_ms": 112.0,
+ "t40_ms": 265.0,
+ "active_s": 0.348,
+ "onset_s": 0.032,
+ "env_oct": 0.251,
+ "curve_db": 4.29,
+ "t40_ratio": 1.082,
+ "attack_ratio": 1.503,
+ "active_ratio": 1.036,
+ "ref_median": {
+ "attack_ms": 74.5,
+ "t40_ms": 245.0,
+ "active_s": 0.336
+ },
+ "phrase_z": 4.03,
+ "phrase_min": 3.94,
+ "margin": 2.09,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.093
+ ],
+ [
+ "Sound effect",
+ 0.04
+ ],
+ [
+ "Music",
+ 0.024
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.093,
+ "floor_db": -28.9,
+ "stereo_corr": 0.996,
+ "darken": -0.2,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -67.1,
+ "first_sample": 0.0,
+ "tail_db": -47.8,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "real_rev": [
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/bench/fs-352719_1.wav",
+ "length_s": 0.435,
+ "typ_z": 1.44,
+ "nn_cos": 0.897,
+ "tone_db": 7.49,
+ "tone_ratio": 0.99,
+ "n_outside": 2,
+ "outside": [
+ [
+ "157 Hz",
+ -8.1,
+ "below"
+ ],
+ [
+ "2.52 kHz",
+ -6.5,
+ "below"
+ ]
+ ],
+ "attack_ms": 41.0,
+ "t40_ms": 220.0,
+ "active_s": 0.325,
+ "onset_s": 0.065,
+ "env_oct": 0.514,
+ "curve_db": 15.4,
+ "t40_ratio": 1.375,
+ "attack_ratio": 1.323,
+ "active_ratio": 1.601,
+ "ref_median": {
+ "attack_ms": 31.0,
+ "t40_ms": 160.0,
+ "active_s": 0.203
+ },
+ "phrase_z": 4.44,
+ "phrase_min": 4.24,
+ "margin": 3.67,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.102
+ ],
+ [
+ "Sound effect",
+ 0.08
+ ],
+ [
+ "Zipper (clothing)",
+ 0.035
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.102,
+ "floor_db": -44.6,
+ "stereo_corr": 1.0,
+ "darken": -0.45,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -54.7,
+ "first_sample": 0.0,
+ "tail_db": -64.7,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/bench/fs-216675_1.wav",
+ "length_s": 0.23,
+ "typ_z": 2.39,
+ "nn_cos": 0.926,
+ "tone_db": 8.07,
+ "tone_ratio": 1.07,
+ "n_outside": 4,
+ "outside": [
+ [
+ "50 Hz",
+ -29.1,
+ "below"
+ ],
+ [
+ "39 Hz",
+ -14.3,
+ "below"
+ ],
+ [
+ "2.52 kHz",
+ -5.3,
+ "below"
+ ]
+ ],
+ "attack_ms": 33.0,
+ "t40_ms": 110.0,
+ "active_s": 0.177,
+ "onset_s": 0.023,
+ "env_oct": 0.276,
+ "curve_db": 7.54,
+ "t40_ratio": 0.688,
+ "attack_ratio": 1.065,
+ "active_ratio": 0.872,
+ "ref_median": {
+ "attack_ms": 31.0,
+ "t40_ms": 160.0,
+ "active_s": 0.203
+ },
+ "phrase_z": 4.34,
+ "phrase_min": 3.89,
+ "margin": 3.35,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Music",
+ 0.209
+ ],
+ [
+ "Sound effect",
+ 0.206
+ ],
+ [
+ "Bang",
+ 0.138
+ ]
+ ],
+ "as_names": false,
+ "as_p": 0.005,
+ "floor_db": -46.7,
+ "stereo_corr": 1.0,
+ "darken": -0.77,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -98.6,
+ "first_sample": 0.0,
+ "tail_db": -63.8,
+ "lead_s": 0.001,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/bench/fs-463763_1.wav",
+ "length_s": 0.435,
+ "typ_z": -6.16,
+ "nn_cos": 0.698,
+ "tone_db": 8.45,
+ "tone_ratio": 1.12,
+ "n_outside": 10,
+ "outside": [
+ [
+ "1 kHz",
+ 20.5,
+ "above"
+ ],
+ [
+ "1.26 kHz",
+ 14.9,
+ "above"
+ ],
+ [
+ "794 Hz",
+ 14.7,
+ "above"
+ ]
+ ],
+ "attack_ms": 117.0,
+ "t40_ms": 295.0,
+ "active_s": 0.406,
+ "onset_s": 0.009,
+ "env_oct": 1.266,
+ "curve_db": 23.08,
+ "t40_ratio": 1.844,
+ "attack_ratio": 3.774,
+ "active_ratio": 2.0,
+ "ref_median": {
+ "attack_ms": 31.0,
+ "t40_ms": 160.0,
+ "active_s": 0.203
+ },
+ "phrase_z": 3.92,
+ "phrase_min": 3.86,
+ "margin": 2.32,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.091
+ ],
+ [
+ "Whip",
+ 0.073
+ ],
+ [
+ "Sound effect",
+ 0.063
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.091,
+ "floor_db": -32.2,
+ "stereo_corr": 0.995,
+ "darken": -0.03,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -63.4,
+ "first_sample": 0.0,
+ "tail_db": -49.2,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/bench/fs-463763_2.wav",
+ "length_s": 0.398,
+ "typ_z": -3.76,
+ "nn_cos": 0.745,
+ "tone_db": 8.15,
+ "tone_ratio": 1.08,
+ "n_outside": 9,
+ "outside": [
+ [
+ "1 kHz",
+ 20.6,
+ "above"
+ ],
+ [
+ "794 Hz",
+ 14.8,
+ "above"
+ ],
+ [
+ "10.1 kHz",
+ -13.7,
+ "below"
+ ]
+ ],
+ "attack_ms": 108.0,
+ "t40_ms": 270.0,
+ "active_s": 0.347,
+ "onset_s": 0.033,
+ "env_oct": 1.11,
+ "curve_db": 22.91,
+ "t40_ratio": 1.688,
+ "attack_ratio": 3.484,
+ "active_ratio": 1.709,
+ "ref_median": {
+ "attack_ms": 31.0,
+ "t40_ms": 160.0,
+ "active_s": 0.203
+ },
+ "phrase_z": 4.14,
+ "phrase_min": 4.01,
+ "margin": 2.62,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.138
+ ],
+ [
+ "Sound effect",
+ 0.063
+ ],
+ [
+ "Music",
+ 0.033
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.138,
+ "floor_db": -28.7,
+ "stereo_corr": 0.991,
+ "darken": -0.18,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -51.9,
+ "first_sample": 0.0,
+ "tail_db": -49.3,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ },
+ {
+ "file": "E:/earshot/out/refs/whoosh_fast/bench/fs-463763_3.wav",
+ "length_s": 0.392,
+ "typ_z": -4.94,
+ "nn_cos": 0.721,
+ "tone_db": 8.21,
+ "tone_ratio": 1.09,
+ "n_outside": 9,
+ "outside": [
+ [
+ "1 kHz",
+ 21.5,
+ "above"
+ ],
+ [
+ "794 Hz",
+ 15.0,
+ "above"
+ ],
+ [
+ "10.1 kHz",
+ -12.9,
+ "below"
+ ]
+ ],
+ "attack_ms": 112.0,
+ "t40_ms": 265.0,
+ "active_s": 0.348,
+ "onset_s": 0.032,
+ "env_oct": 1.12,
+ "curve_db": 21.43,
+ "t40_ratio": 1.656,
+ "attack_ratio": 3.613,
+ "active_ratio": 1.714,
+ "ref_median": {
+ "attack_ms": 31.0,
+ "t40_ms": 160.0,
+ "active_s": 0.203
+ },
+ "phrase_z": 4.03,
+ "phrase_min": 3.94,
+ "margin": 2.09,
+ "wins": 1.0,
+ "best_confusion": "a steam hiss",
+ "as_top": [
+ [
+ "Whoosh, swoosh, swish",
+ 0.093
+ ],
+ [
+ "Sound effect",
+ 0.04
+ ],
+ [
+ "Music",
+ 0.024
+ ]
+ ],
+ "as_names": true,
+ "as_p": 0.093,
+ "floor_db": -28.9,
+ "stereo_corr": 0.996,
+ "darken": -0.2,
+ "artifacts": [],
+ "artifact_detail": {
+ "clipped_samples": 0,
+ "dc_db": -67.1,
+ "first_sample": 0.0,
+ "tail_db": -47.8,
+ "lead_s": 0.0,
+ "peak_dbfs": -1.0
+ }
+ }
+ ],
+ "diversity_bench": {
+ "n": 5,
+ "cos_mean": 0.772,
+ "cos_max": 0.951,
+ "tone_min_db": 2.28,
+ "tone_mean_db": 8.01,
+ "near_duplicates": []
+ },
+ "diversity_forge": {
+ "n": 5,
+ "cos_mean": 0.826,
+ "cos_max": 0.892,
+ "tone_min_db": 5.21,
+ "tone_mean_db": 10.47,
+ "near_duplicates": []
+ },
+ "inside": {
+ "per_sound": [
+ 7,
+ 7,
+ 7
+ ],
+ "n_checks": 7,
+ "limits": {
+ "typ_z": -0.874,
+ "tone_db": 10.348,
+ "env_oct": 1.317,
+ "curve_db": 21.472,
+ "phrase_z": 3.88,
+ "margin": 2.314,
+ "as_p": 0.007
+ },
+ "real_per_sound": [
+ 1,
+ 7,
+ 7,
+ 4,
+ 6,
+ 7,
+ 4,
+ 7,
+ 7,
+ 6
+ ]
+ },
+ "summary": {
+ "made": {
+ "n": 3,
+ "typ_z": 2.05,
+ "tone_db": 8.58,
+ "env_oct": 0.89,
+ "curve_db": 15.93,
+ "phrase_z": 4.46,
+ "margin": 3.31,
+ "as_p": 0.02,
+ "tone_ratio": 1.87,
+ "t40_ratio": 0.51,
+ "wins": 1.0,
+ "floor_db": -43.9,
+ "stereo_corr": 0.9,
+ "darken": -0.32,
+ "as_names": 0.33,
+ "artifacts": 0,
+ "var_cos": 0.957
+ },
+ "real_cross": {
+ "n": 5,
+ "typ_z": 0.57,
+ "tone_db": 10.31,
+ "env_oct": 1.22,
+ "curve_db": 20.2,
+ "phrase_z": 4.14,
+ "margin": 2.78,
+ "as_p": 0.06,
+ "tone_ratio": 2.25,
+ "t40_ratio": 0.6,
+ "wins": 1.0,
+ "floor_db": -65.4,
+ "stereo_corr": 0.81,
+ "darken": -0.16,
+ "as_names": 0.4,
+ "artifacts": 0,
+ "var_cos": 0.826
+ },
+ "real_loo": {
+ "n": 5,
+ "typ_z": 1.52,
+ "tone_db": 4.58,
+ "env_oct": 0.4,
+ "curve_db": 8.21,
+ "phrase_z": 4.14,
+ "margin": 2.62,
+ "as_p": 0.09,
+ "tone_ratio": 0.66,
+ "t40_ratio": 1.08,
+ "wins": 1.0,
+ "floor_db": -32.2,
+ "stereo_corr": 1.0,
+ "darken": -0.2,
+ "as_names": 0.8,
+ "artifacts": 0,
+ "var_cos": 0.772
+ }
+ }
+ }
+ },
+ "seconds": 11.5
+}
\ No newline at end of file
diff --git a/eval/forge/results_v1.md b/eval/forge/results_v1.md
new file mode 100644
index 0000000..7542249
--- /dev/null
+++ b/eval/forge/results_v1.md
@@ -0,0 +1,50 @@
+# forge benchmark: v1
+
+made: ['typing_key=demo/forge-test/keypress/final/*.wav', 'click=demo/forge-test/click/final/*.wav', 'pop=demo/forge-test/pop/final/*.wav', 'whoosh_fast=demo/forge-test/whoosh/final/*.wav', 'impact=demo/forge-test/impact/final/*.wav']; refs: E:\earshot\.cache\refsets.json; 2026-10-16 04:11
+
+Medians per set. made = the made sounds vs the bench refs; real x = the forge refs (real, never in the bench) vs the bench refs; real loo = each bench ref vs the other bench refs. typ z, phrase z, margin, AS names: higher is better; tone dB, env oct (octaves off in attack / t40 / length), curve dB: lower is better; t40 x = tail length / bench median (1 = same). inside = checks (of n) a made sound lands inside the real 80% limit, median over the made sounds (real = the same count for the real rows). var cos = mean pairwise CLAP cosine within the set (made variants / forge refs / bench refs): high = the variants are near clones.
+
+| category | set | n | typ z | tone dB | env oct | curve dB | t40 x | phrase z | margin | AS names | artifacts | var cos | inside |
+|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
+| click | made | 3 | 1.89 | 7.6 | 0.56 | 8.85 | 0.56 | 3.12 | 1.13 | 0 | 0 | 0.886 | 7 / 7 (real 6) |
+| | real x | 8 | 0.16 | 11.33 | 0.22 | 9.92 | 1.33 | 2.77 | 0.53 | 0.12 | 0 | 0.651 | |
+| | real loo | 7 | -0.1 | 9.85 | 0.46 | 13.19 | 0.86 | 2.67 | 0.27 | 0 | 0 | 0.576 | |
+| impact | made | 3 | | 6.7 | 1.54 | 24.86 | 0.5 | 3.68 | 1.81 | 1 | 0 | 0.931 | 5 / 6 (real 3) |
+| | real x | 2 | | 6.19 | 2.5 | 31.98 | 0.46 | 3.04 | 1.09 | 0.5 | 0 | 0.659 | |
+| | note: thin: the library has three cinematic hits (all three were the v1 impact forge refs); one is held out as bench, so v1 impact bench numbers are optimistic. | | | | | | | | | | | | |
+| pop | made | 3 | 1.33 | 8.59 | 0.68 | 6.29 | 1 | 3.09 | -0.1 | 1 | 0 | 0.917 | 7 / 7 (real 6) |
+| | real x | 7 | -0.09 | 16.06 | 1.51 | 20.54 | 0.91 | 2.46 | -0.5 | 0.29 | 0 | 0.501 | |
+| | real loo | 6 | 1.13 | 13.66 | 0.41 | 12.5 | 1.02 | 1.96 | -1.06 | 0.67 | 0 | 0.522 | |
+| typing_key | made | 3 | 1.2 | 8.2 | 1.07 | 14.32 | 1.67 | 5.56 | 3.57 | 0.33 | 0 | 0.965 | 7 / 7 (real 6) |
+| | real x | 15 | 0.26 | 8.88 | 1.01 | 12.2 | 1.29 | 4.75 | 2.83 | 0.53 | 0 | 0.721 | |
+| | real loo | 12 | 0.23 | 5.47 | 1.52 | 16.7 | 1 | 3.71 | 1.53 | 0.42 | 0 | 0.673 | |
+| whoosh_fast | made | 3 | 2.05 | 8.58 | 0.89 | 15.93 | 0.51 | 4.46 | 3.31 | 0.33 | 0 | 0.957 | 7 / 7 (real 6.5) |
+| | real x | 5 | 0.57 | 10.31 | 1.22 | 20.2 | 0.6 | 4.14 | 2.78 | 0.4 | 0 | 0.826 | |
+| | real loo | 5 | 1.52 | 4.58 | 0.4 | 8.21 | 1.08 | 4.14 | 2.62 | 0.8 | 0 | 0.772 | |
+
+Texture (medians; model-free, not in any forge objective): floor dB = 10th percentile 5 ms frame re the loudest (-120 = digital silence), stereo corr (1 = mono), darken = log2 centroid at half the tail / at the peak (< 0: the decay darkens).
+
+| category | set | floor dB | stereo corr | darken |
+|---|---|---|---|---|
+| click | made | -49.0 | 1.0 | 0.41 |
+| | real x | -42.85 | 0.99 | -0.12 |
+| | real loo | -44.8 | 1.0 | -0.07 |
+| impact | made | -64.7 | 0.41 | -1.61 |
+| | real x | -70.55 | 0.54 | -1.3 |
+| pop | made | -65.0 | 1.0 | 0.19 |
+| | real x | -53.2 | 1.0 | 0.21 |
+| | real loo | -47.2 | 1.0 | 0.12 |
+| typing_key | made | -54.7 | 0.94 | 0.4 |
+| | real x | -43.7 | 0.76 | 0.04 |
+| | real loo | -39.65 | 1.0 | 0.09 |
+| whoosh_fast | made | -43.9 | 0.9 | -0.32 |
+| | real x | -65.4 | 0.81 | -0.16 |
+| | real loo | -32.2 | 1.0 | -0.2 |
+
+Diversity among the made variants (CLAP cosine mean / max, tonal distance min dB, near duplicates):
+
+- click: cos 0.886 / 0.93, tone min 7.24 dB, 0 near duplicate pair(s) (bench refs: cos 0.576 / 0.792, tone min 6.13 dB)
+- impact: cos 0.931 / 0.943, tone min 3.76 dB, 0 near duplicate pair(s)
+- pop: cos 0.917 / 0.96, tone min 4.49 dB, 0 near duplicate pair(s) (bench refs: cos 0.522 / 0.832, tone min 7.62 dB)
+- typing_key: cos 0.965 / 0.967, tone min 4.27 dB, 0 near duplicate pair(s) (bench refs: cos 0.673 / 0.919, tone min 4.64 dB)
+- whoosh_fast: cos 0.957 / 0.973, tone min 4.11 dB, 0 near duplicate pair(s) (bench refs: cos 0.772 / 0.951, tone min 2.28 dB)
diff --git a/tests/test_forge_critic.py b/tests/test_forge_critic.py
new file mode 100644
index 0000000..5065d8f
--- /dev/null
+++ b/tests/test_forge_critic.py
@@ -0,0 +1,259 @@
+"""Fast tests for the forge critic v2's model-free ears: every artifact detector and polish measure on synthetic
+signals (a phase-jump click, a naive vs band-limited oscillator, a resonant ring, a comb-filtered tail, a truncated
+tail, digital silence, clipping, a slow attack, a bumpy tail ...), the spectro-temporal envelope match (EnvRef),
+PolishRef's ref-relative tolerances, and the v2 total. No models are loaded."""
+import numpy as np
+import pytest
+from scipy import signal
+
+from earshot.forge import polish as P
+from earshot.forge.critic import W2, resolve_mode, total_v2
+from earshot.forge.tone import CEN, EnvRef, bandgram, env_dist
+
+SR = 48000
+
+
+def t_(s):
+ return np.arange(int(s * SR)) / SR
+
+
+def decay_sine(f=440.0, tau=0.1, dur=0.6, fade=0.001):
+ t = t_(dur)
+ x = np.sin(2 * np.pi * f * t) * np.exp(-t / tau)
+ n = int(fade * SR)
+ x[:n] *= np.linspace(0, 1, n)
+ return x
+
+
+def burst(dur=0.6, tau=0.12, seed=0, floor=1e-4):
+ r = np.random.default_rng(seed)
+ t = t_(dur)
+ x = r.standard_normal(len(t)) * np.exp(-t / tau)
+ x[:48] *= np.linspace(0, 1, 48)
+ return x + floor * r.standard_normal(len(t))
+
+
+def saw(f0=440.0, dur=1.0, bandlimited=False):
+ t = t_(dur)
+ ph = f0 * t
+ if bandlimited:
+ x = sum(np.sin(2 * np.pi * h * ph) / h for h in range(1, int(20000 / f0) + 1)) * 2 / np.pi
+ else:
+ x = 2 * (ph % 1.0) - 1
+ return 0.5 * x * np.minimum(1, t / 0.01) * np.minimum(1, (dur - t) / 0.05)
+
+
+# ------------------------------------------------------------------------------------------------ artifacts
+def test_clicks_finds_a_phase_jump_not_a_clean_decay():
+ x = decay_sine()
+ y = x.copy()
+ y[int(0.3013 * SR):] *= -1 # a jump off a zero crossing
+ assert P.clicks(x)['clicks'] == 0
+ c = P.clicks(y)
+ assert c['clicks'] == 1 and abs(c['click_times'][0] - 0.301) < 0.003
+ assert P.clicks(x)['edge_db'] < -60
+ cut = decay_sine(tau=1.0, dur=0.2) # ends while loud: the file jumps to zero
+ assert P.clicks(cut)['edge_db'] > -40
+
+
+def test_clicks_ignores_a_physical_tick():
+ x = burst(dur=0.4, tau=0.05)
+ tick = decay_sine(5000, 0.0004, 0.004) * 0.05 # a 0.4 ms tick at 200 ms spreads over many samples
+ x[int(0.2 * SR):int(0.2 * SR) + len(tick)] += tick
+ assert P.clicks(x)['clicks'] == 0
+
+
+def test_aliasing_naive_vs_bandlimited_oscillator():
+ a, b = P.aliasing(saw()), P.aliasing(saw(bandlimited=True))
+ assert a['fold'] > 0.05 and b['fold'] < 0.01
+ assert a['nyq_db'] > -30 and b['nyq_db'] < -40
+ n = P.aliasing(burst())
+ assert n['fold'] == 0.0 and n['harmonic_frames'] == 0 # noise has no harmonic grid: not judged
+
+
+def test_dc_offset_and_infrasound():
+ x = burst()
+ assert P.dc(x)['dc_db'] < -60
+ assert P.dc(x + 0.05)['dc_db'] > -40 and P.dc(x + 0.05)['infra_db'] > -20
+
+
+def test_resonance_finds_a_ring_in_noise():
+ x = burst(dur=0.5, tau=0.1)
+ r0 = P.resonances(x)
+ r1 = P.resonances(x + 0.5 * decay_sine(3000, 0.1, 0.5))
+ assert r0['res_db'] < 10
+ assert r1['res_db'] > 15 and abs(r1['res_hz'] - 3000) < 200
+
+
+def test_evolution_frozen_vs_sweeping():
+ r = np.random.default_rng(1)
+ t = t_(0.8)
+ env = np.sin(np.pi * t / 0.8)
+ frozen = r.standard_normal(len(t)) * env
+ w = r.standard_normal(len(t))
+ sos_lo = signal.butter(4, 400, 'lowpass', fs=SR, output='sos')
+ sos_hi = signal.butter(4, 6000, 'highpass', fs=SR, output='sos')
+ mix = np.clip((t - 0.2) / 0.4, 0, 1) # crossfade from a dark to a bright noise
+ moving = (signal.sosfilt(sos_lo, w) * (1 - mix) * 4 + signal.sosfilt(sos_hi, w) * mix) * env
+ ef, em = P.evolution(frozen), P.evolution(moving)
+ assert em['evolution_db'] > 2 * ef['evolution_db']
+ assert em['centroid_oct'] > ef['centroid_oct'] + 1
+
+
+def test_ring_comb_tail_vs_natural_tail():
+ natural = burst(dur=1.5, tau=0.25)
+ b = burst(dur=0.1, tau=0.02)
+ d = int(0.005 * SR)
+ a = np.zeros(d + 1)
+ a[0], a[d] = 1.0, -0.9
+ comb = signal.lfilter([1.0], a, np.concatenate([b, np.zeros(SR)]))
+ assert P.ring(natural)['ring_z'] < 8
+ c = P.ring(comb)
+ assert c['ring_z'] > 20 and abs(c['ring_ms'] - 5.0) < 0.3
+
+
+def test_truncated_tail():
+ assert P.truncated(burst(dur=2.0))['trunc_db'] is None
+ t = P.truncated(burst(dur=0.3))
+ assert t['trunc_db'] is not None and t['trunc_db'] > -30
+ faded = burst(dur=0.3)
+ faded[-480:] *= np.linspace(1, 0, 480) # a 10 ms fade does not hide a cut tail
+ assert P.truncated(faded)['trunc_db'] is not None
+
+
+def test_digital_silence_inside():
+ x = burst()
+ gap = np.concatenate([x[:4800], np.zeros(4800), x[4800:9600]])
+ assert abs(P.silence(gap)['silence_ms'] - 100) < 2
+ assert P.silence(x)['silence_ms'] == 0.0
+
+
+def test_clipping_flat_tops():
+ s = np.sin(2 * np.pi * 200 * t_(0.3))
+ assert P.clipping(np.clip(2 * s, -1, 1))['clip_runs'] > 10
+ assert P.clipping(s)['clip_runs'] == 0
+ assert P.clipping(np.stack([s, 1.5 * s], 1))['overs'] > 0
+
+
+def test_attack_tail_body_and_loudness():
+ x = burst()
+ slow = x.copy()
+ slow[:2400] *= np.linspace(0, 1, 2400)
+ assert P.attack(slow)['attack_ms'] > 20 > P.attack(x)['attack_ms']
+ bumpy = x.copy()
+ bumpy[9600:] += 0.6 * burst(dur=0.4, tau=0.05, seed=3)
+ assert P.tail(bumpy)['bump_db'] > 4 > P.tail(x)['bump_db']
+ assert P.tail(x)['decay_db_s'] < -40
+ assert P.body(decay_sine(2000, 0.003, 0.1))['crest_db'] > P.body(x)['crest_db'] + 2
+ ls = P.loudness_spread([x, 0.5 * x])
+ assert abs(ls['spread_lu'] - 6.02) < 0.05
+
+
+def test_measure_has_every_detector_and_plain_data():
+ m = P.measure(burst())
+ for k in ('clicks', 'edge_db', 'fold', 'nyq_db', 'dc_db', 'infra_db', 'res_db', 'ring_z', 'trunc_db', 'silence_ms',
+ 'clip_runs', 'evolution_db', 'attack_ms', 'onset_hf_db', 'bump_db', 't40_ms', 'crest_db'):
+ assert k in m
+ assert all(isinstance(v, (int, float, str, list, type(None))) for v in m.values())
+
+
+# ------------------------------------------------------------------------------------------------ PolishRef
+def test_polishref_flags_hard_defects_and_passes_clean():
+ refs = [decay_sine(f, 0.12, 0.8) for f in (400, 500, 630)]
+ pr = P.PolishRef(refs)
+ clean = pr.check(decay_sine(520, 0.12, 0.8))
+ assert clean['art_pen'] == 0 and clean['hard'] == [] and clean['words'].startswith('clean')
+ y = decay_sine(520, 0.12, 0.8)
+ y[int(0.2013 * SR):] *= -1
+ c = pr.check(y)
+ assert 'clicks' in c['hard'] and c['art_pen'] >= 1 and 'click' in c['words']
+ cut = pr.check(decay_sine(520, 0.12, 0.25))
+ assert 'trunc_db' in cut['hard']
+ clip = pr.check(np.clip(3 * decay_sine(520, 0.12, 0.8), -1, 1))
+ assert 'clip_runs' in clip['hard']
+
+
+def test_polishref_refs_loosen_tolerances_and_designed_softens():
+ refs = [burst(seed=s) + 0.1 for s in range(3)] # every ref carries the same offset: not a defect here
+ pr = P.PolishRef(refs)
+ assert 'dc_db' not in pr.check(burst(seed=7) + 0.1)['art']
+ assert 'dc_db' in P.PolishRef([]).check(burst(seed=7) + 0.1)['hard']
+ y = decay_sine(520, 0.12, 0.8)
+ y[int(0.2013 * SR):] *= -1
+ d = P.PolishRef([], designed=True).check(y)
+ assert 'clicks' not in d['hard'] and d['art'].get('clicks', 0) > 0
+
+
+def test_polishref_relative_polish_and_leave_one_out():
+ refs = [burst(seed=s, tau=0.1) for s in range(4)]
+ pr = P.PolishRef(refs)
+ slow = burst(seed=8, tau=0.1)
+ slow[:4800] *= np.linspace(0, 1, 4800)
+ assert pr.check(slow)['polish'].get('attack_ms', 0) > 0
+ r = pr.check(None, exclude=0, m=pr.M[0])
+ assert r['polish_pen'] == 0
+
+
+# ------------------------------------------------------------------------------------------------ envelope
+def test_bandgram_shape_and_level_independence():
+ x = burst()
+ G = bandgram(x)
+ assert G.shape == (int(np.ceil(len(x) / 480)), len(CEN))
+ tot = 10 * np.log10((10 ** (G / 10)).sum(1))
+ assert abs(tot.max()) < 0.01
+ assert np.allclose(G, bandgram(0.1 * x), atol=1e-6)
+
+
+def test_env_dist_tolerates_small_stretch_but_not_a_cut_tail():
+ x = burst(dur=1.0, tau=0.15)
+ G = bandgram(x)
+ st = signal.resample(x, int(len(x) * 1.05)) # 5 % slower (and 5 % lower): inside the tolerance
+ cut = x.copy()
+ cut[int(0.25 * SR):] *= np.linspace(1, 0, len(cut) - int(0.25 * SR)) ** 8
+ assert env_dist(G, G) < 0.5
+ assert env_dist(bandgram(st), G) < 3.0
+ assert env_dist(bandgram(cut), G) > env_dist(bandgram(st), G) + 2
+
+
+def test_envref_nearest_ratio_and_outside_cells():
+ refs = [burst(dur=0.8, tau=tau, seed=s) for s, tau in enumerate((0.08, 0.1, 0.12, 0.15))]
+ er = EnvRef(refs)
+ m = er.check(refs[1])
+ assert m['env_db'] < 0.5 and m['nearest_i'] == 1
+ loo = er.check(None, exclude=1, G=er.G[1])
+ assert loo['nearest_i'] != 1 and loo['env_ratio'] is not None
+ assert EnvRef.penalty(loo) < 0.5
+ odd = burst(dur=0.8, tau=0.1, seed=7) + 2.0 * decay_sine(3000, 0.3, 0.8) # a long 3 kHz ring the refs lack
+ o = er.check(odd)
+ assert o['outside'] and any('1.6-3.2 kHz' in k for k, _, _ in o['outside'])
+ assert o['env_out'] > 0 and EnvRef.penalty(o) > 0.5
+ assert 'beyond every ref' in o['words']
+
+
+# ------------------------------------------------------------------------------------------------ the v2 total
+def _row(**kw):
+ r = {'match_z': 4.0, 'alt_margin': 1.0, 'p_real': 0.9, 'ref_typ_z': 0.5, 'feat_pen': 0.2, 'tone_pen': 0.0,
+ 'env_pen': 0.0, 'art_pen': 0.0, 'polish_pen': 0.0, 'hard': []}
+ r.update(kw)
+ return r
+
+
+def test_total_v2_arithmetic_and_modes():
+ t, ok = total_v2(_row())
+ assert abs(t - (4.0 + 2.0 * 0.8 + 0.5 * 0.5 - 0.7 * 0.2)) < 1e-9 and ok
+ t2, _ = total_v2(_row(tone_pen=1.0, env_pen=0.5, art_pen=1.0, polish_pen=1.0))
+ assert abs((t - t2) - (W2['tone'] + 0.5 * W2['env'] + W2['art'] + W2['polish'])) < 1e-9
+ td, okd = total_v2(_row(p_real=0.1), mode='designed') # realness neither counts nor gates in designed mode
+ assert abs(td - (4.0 + 0.25 - 0.14)) < 1e-9 and okd
+ assert not total_v2(_row(p_real=0.1))[1]
+ assert not total_v2(_row(hard=['clicks']))[1]
+ assert not total_v2(_row(env_pen=2.0))[1]
+ assert total_v2(_row(p_real=0.1), W={'real': 0.0})[0] == pytest.approx(td, abs=1e-9)
+
+
+def test_resolve_mode():
+ assert resolve_mode('auto', 'a key press', []) == 'designed'
+ assert resolve_mode('auto', 'a glitchy digital stutter', ['r.wav']) == 'designed'
+ assert resolve_mode('auto', 'a deep sub drop', ['r.wav']) == 'designed'
+ assert resolve_mode('auto', 'a single key press', ['r.wav']) == 'real'
+ assert resolve_mode('real', 'a glitch', []) == 'real'
diff --git a/tests/test_forge_dsp.py b/tests/test_forge_dsp.py
new file mode 100644
index 0000000..dfc9b7e
--- /dev/null
+++ b/tests/test_forge_dsp.py
@@ -0,0 +1,420 @@
+"""Fast tests for the forge sound toolkit v2 (earshot/forge/dsp.py sources + layering, earshot/forge/fx.py effects):
+shape, finite, level (no clipping unless asked), determinism and that each tool does what it claims, measured on
+synthetic signals. No models are loaded."""
+import numpy as np
+import pytest
+from scipy import ndimage, signal
+
+from earshot.forge import dsp, fx, tone
+
+SR = 48000
+
+
+def ok(y, peak=1.0):
+ """finite, non-silent, and no sample above `peak` (1.0 = 0 dBFS)"""
+ y = np.asarray(y)
+ assert np.isfinite(y).all()
+ assert np.abs(y).max() > 1e-6
+ assert np.abs(y).max() <= peak + 1e-9, np.abs(y).max()
+ return y
+
+
+def same(f):
+ """f(rng) twice with the same seed gives identical samples"""
+ a, b = f(dsp.rng(5)), f(dsp.rng(5))
+ assert np.array_equal(a, b)
+ return a
+
+
+def peak_hz(x, lo=20, hi=20000):
+ F = np.abs(np.fft.rfft(x * np.hanning(len(x))))
+ f = np.fft.rfftfreq(len(x), 1 / SR)
+ s = (f >= lo) & (f <= hi)
+ return float(f[s][np.argmax(F[s])])
+
+
+def ring_stats(ir, rt60, t1=0.1, band=(100, 6000)):
+ """decay-compensated late tail: 99.9th percentile (dB) of the long-FFT spectrum over its local mean. Gaussian
+ noise (a real late field) sits near 8 dB; resolvable modes (metallic ringing) push it to 11-16 dB"""
+ m = ir if ir.ndim == 1 else ir[:, 0]
+ seg = m[int(t1 * SR):int((t1 + 1.5 * rt60) * SR)]
+ seg = seg * np.exp(6.91 * (np.arange(len(seg)) / SR + t1) / rt60)
+ P = np.abs(np.fft.rfft(seg * np.hanning(len(seg)))) ** 2
+ f = np.fft.rfftfreq(len(seg), 1 / SR)
+ sm = ndimage.uniform_filter1d(P, max(9, int(60 / f[1])))
+ s = (f > band[0]) & (f < band[1])
+ return float(np.percentile(10 * np.log10(P[s] / sm[s]), 99.9))
+
+
+def band_rt60(x, lo, hi):
+ b = dsp.butter(x, 'band', (lo, hi), order=3)
+ e = np.cumsum((b ** 2)[::-1])[::-1]
+ e = 10 * np.log10(e / e[0] + 1e-30)
+ return (np.argmax(e < -25) - np.argmax(e < -5)) / SR * 3
+
+
+IMP = np.eye(1, 4800, 0)[0]
+
+
+# ------------------------------------------------------------------------------------------------ dsp: sources
+def test_curve_and_sweep():
+ n = dsp.n_of(1.0)
+ assert np.allclose(dsp.curve(3.0, 10), 3.0)
+ c = dsp.curve([(0, 0.0), (1.0, 1.0)], n)
+ assert len(c) == n and abs(c[n // 2] - 0.5) < 0.01
+ assert len(dsp.curve(np.array([0.0, 1.0]), 100)) == 100
+ f = dsp.sweep(100, 1600, 1.0)
+ assert abs(f[0] - 100) < 1e-9 and abs(f[-1] - 1600) < 1e-6
+ assert abs(f[n // 2] - 400) < 1 # exponential: half time = half the octaves
+ e = dsp.sweep(100, 1600, 1.0, 'ease', power=2.0)
+ assert e[n // 2] < f[n // 2] and np.all(np.diff(e) >= 0)
+
+
[email protected]('shape', ['sine', 'saw', 'square', 'tri'])
+def test_osc_shapes(shape):
+ y = dsp.osc(440, shape, 0.5)
+ assert len(y) == SR // 2 and np.isfinite(y).all()
+ assert 0.9 < np.abs(y).max() < 1.15
+ assert abs(peak_hz(y) - 440) < 3
+
+
+def test_osc_saw_is_band_limited():
+ """a 5 kHz saw: harmonics land on 5, 10, 15, 20 kHz; everything else would be aliasing"""
+ y = dsp.osc(5000, 'saw', 1.0)
+ F = np.abs(np.fft.rfft(y * np.hanning(len(y)))) ** 2
+ f = np.fft.rfftfreq(len(y), 1 / SR)
+ harm = np.zeros(len(f), bool)
+ for k in range(1, 5):
+ harm |= np.abs(f - 5000 * k) < 20
+ alias = 10 * np.log10(F[~harm & (f > 30)].sum() / F[harm].sum())
+ naive = 2 * ((5000 * np.arange(SR) / SR) % 1.0) - 1
+ Fn = np.abs(np.fft.rfft(naive * np.hanning(SR))) ** 2
+ alias_naive = 10 * np.log10(Fn[~harm & (f > 30)].sum() / Fn[harm].sum())
+ assert alias < alias_naive - 8 and alias < -20
+
+
+def test_shepard_unit_rms_deterministic():
+ for kind in ('tone', 'noise'):
+ y = same(lambda r: dsp.shepard(1.0, r, kind=kind))
+ assert abs(np.sqrt(np.mean(y ** 2)) - 1) < 1e-6 and np.isfinite(y).all()
+ up, dn = dsp.shepard(1.0, dsp.rng(1), up=True), dsp.shepard(1.0, dsp.rng(1), up=False)
+ assert not np.allclose(up, dn)
+
+
+def test_shepard_voices_glide_in_the_right_direction():
+ """a single voice region: the strongest frequency inside 400-800 Hz moves up for up=True, down for up=False"""
+ for up, sgn in ((True, 1), (False, -1)):
+ y = dsp.shepard(2.0, dsp.rng(2), up=up, rate=0.5, f_lo=50, f_hi=12800)
+ a, b = peak_hz(y[:8192], 400, 800), peak_hz(y[SR // 4:SR // 4 + 8192], 400, 800)
+ d = np.log2(b / a) % 1.0 # octave-wrapped move of the nearest voice
+ assert (d < 0.5) == (sgn > 0)
+
+
+def test_wander_and_air():
+ c = same(lambda r: dsp.wander(2.0, r, rate=5))
+ assert abs(c.mean()) < 1e-6 and abs(c.std() - 1) < 1e-6 and len(c) == 2 * SR
+ assert np.abs(np.diff(c)).max() < 0.01 # smooth (a control signal, not audio noise)
+ y = same(lambda r: dsp.air(1.5, r, center=1000, flutter=0.0, bright=0.0))
+ assert abs(np.sqrt(np.mean(y ** 2)) - 1) < 1e-6
+ f, P = signal.welch(y, SR, nperseg=8192)
+ psd = lambda hz: 10 * np.log10(np.mean(P[(f > hz / 1.2) & (f < hz * 1.2)]))
+ shape = lambda u: -10 * np.log10(2 ** (-6.0 / 3.0103 * u) + 2 ** (4.5 / 3.0103 * u)) # the documented curve
+ for u in (-2, 2, 3): # rises 6 dB / oct below 1 kHz, falls 4.5 dB / oct above
+ assert abs((psd(1000 * 2 ** u) - psd(1000)) - (shape(u) - shape(0))) < 1.5
+ g = dsp.air(2.0, dsp.rng(3), flutter=0.8, whistle=0.3)
+ env = np.sqrt(ndimage.uniform_filter1d(g ** 2, 2400))
+ assert np.percentile(env, 95) / np.percentile(env, 5) > 2.5 # it gusts
+
+
+def test_sub_drop_pitch_falls_and_peaks_at_minus_1():
+ y = ok(same(lambda r: dsp.sub_drop(1.5, r, f0=150, f1=35, drop=0.2, click=0.0)), 0.8913)
+ assert abs(20 * np.log10(np.abs(y).max()) + 1) < 0.01
+ early, late = peak_hz(y[:4096], 20, 400), peak_hz(y[int(0.8 * SR):int(0.8 * SR) + 16384], 20, 400)
+ assert early > 90 and 30 < late < 45
+
+
+def test_bell_partials_and_decay():
+ y = ok(same(lambda r: dsp.bell(1000, 1.5, r, 'bar', decay=0.6, strike=0.0)), 0.8913)
+ assert abs(peak_hz(y, 500, 1500) - 1000) < 5
+ assert abs(peak_hz(y, 2000, 4000) - 2756) < 15 # the free-free bar's inharmonic second mode
+ e = lambda a, b: np.mean(y[int(a * SR):int(b * SR)] ** 2)
+ assert e(0.6, 0.9) < e(0.0, 0.3) * 0.2
+ hi = dsp.butter(y, 'high', 4000)
+ assert np.mean(hi[int(0.5 * SR):] ** 2) < np.mean(hi[:int(0.2 * SR)] ** 2) * 0.05 # highs die first
+ for name in dsp.BELLS:
+ ok(dsp.bell(440, 0.5, dsp.rng(1), name))
+
+
+def test_fm_inharmonic():
+ y = ok(dsp.fm(500, 1.4, 4, 1.0, decay=0.5, index_decay=0.2), 1.0)
+ F = np.abs(np.fft.rfft(y[:16384] * np.hanning(16384)))
+ f = np.fft.rfftfreq(16384, 1 / SR)
+ side = F[np.abs(f - 700) < 10].max() + F[np.abs(f - 1200) < 10].max() # fc + fm, fc + 2 fm - ...
+ assert side > 0.05 * F.max()
+
+
+# ------------------------------------------------------------------------------------------------ dsp: layering
+def test_onset_align_stack_xfade():
+ x = np.zeros(4800)
+ x[1000:1200] = 1.0
+ assert abs(dsp.onset(x) * SR - 1000) < 1
+ assert abs(dsp.onset(dsp.align(x, 0.05)) - 0.05) < 1 / SR
+ assert abs(dsp.onset(dsp.align(x, 0.0))) < 1 / SR
+ m = dsp.stack([(x, 0.0, 0.0), (x, 0.01, -6.0)])
+ assert m.ndim == 1 and len(m) == 4800 + 480
+ s = dsp.stack([(x, 0.0, 0.0, -1.0), {'x': x, 't': 0.0, 'gain': 0.0, 'pan': 1.0}])
+ assert s.shape == (4800, 2) and abs(s[1100, 0] - s[1100, 1]) < 1e-9 # one hard left + one hard right
+ al = dsp.stack([(x, 0.02, 0.0), (np.r_[np.zeros(300), x], 0.02, 0.0)], onset_db=-24)
+ assert abs(dsp.onset(al) - 0.02) < 1 / SR and al.max() == pytest.approx(2.0) # onsets line up exactly
+ ok(dsp.stack([(dsp.sub_drop(0.5), 0, 0), (dsp.sub_drop(0.5), 0, 0)], peak_db=-1.0), 0.8913)
+ a, b = np.ones(1000), np.ones(2000)
+ lin = dsp.xfade(a, b, 0.005, power=False)
+ assert len(lin) == 3000 - 240 and np.allclose(lin, 1.0)
+ pw = dsp.xfade(np.ones((1000, 2)), np.ones((500, 2)), 0.005)
+ assert pw.shape == (1500 - 240, 2)
+
+
+# ------------------------------------------------------------------------------------------------ fx: reverb
+def test_reverb_shapes_and_determinism():
+ x = dsp.sub_drop(0.3)
+ m = same(lambda r: fx.reverb(x, r, rt60=0.8, predelay=0.02, wet=0.3))
+ assert m.ndim == 1 and len(m) == len(x) + dsp.n_of(0.02) + dsp.n_of(1.2 * 0.8 * 1.2) - 1
+ st = fx.reverb(x, dsp.rng(1), rt60=0.8, stereo=True)
+ assert st.shape[1] == 2 and np.isfinite(st).all()
+ s2 = fx.reverb(np.stack([x, -x], axis=1), dsp.rng(1), rt60=0.5)
+ assert s2.ndim == 2
+ assert np.allclose(fx.reverb(x, dsp.rng(1), wet=0.0, rt60=0.5)[:len(x)], x) # dry path untouched
+ assert np.abs(fx.reverb(x, dsp.rng(1), rt60=0.5, wet=0.2)[-100:]).max() < 1e-3 # faded, no cut-off click
+
+
+def test_reverb_decay_and_unit_energy():
+ """band rt60s follow the design (mid, damped highs, longer bass) and the wet IR has about unit energy"""
+ ir = fx.reverb(IMP, dsp.rng(2), rt60=1.5, damp=0.0, bass=1.0, wet=1.0, dry=0.0, early_db=None, stereo=True,
+ air=0.0)
+ assert 0.7 < (ir ** 2).sum(axis=0).mean() < 1.25
+ assert abs(band_rt60(ir[:, 0], 707, 1414) - 1.5) < 0.2
+ ir = fx.reverb(IMP, dsp.rng(2), rt60=1.5, damp=0.6, bass=1.5, wet=1.0, dry=0.0, stereo=True)
+ mid, hi8, low = band_rt60(ir[:, 0], 707, 1414), band_rt60(ir[:, 0], 5657, 11314), band_rt60(ir[:, 0], 60, 120)
+ assert hi8 < 0.75 * mid and low > 1.2 * mid
+ assert abs(np.corrcoef(ir[:, 0], ir[:, 1])[0, 1]) < 0.3 # a wide, decorrelated stereo tail
+ dry_air = fx.reverb(IMP, dsp.rng(2), rt60=3.0, damp=0.3, wet=1.0, dry=0.0, stereo=True, air=0.0)
+ wet_air = fx.reverb(IMP, dsp.rng(2), rt60=3.0, damp=0.3, wet=1.0, dry=0.0, stereo=True, air=1.0)
+ assert band_rt60(wet_air[:, 0], 8500, 17000) < 0.45 * band_rt60(dry_air[:, 0], 8500, 17000) # air absorption
+ assert band_rt60(wet_air[:, 0], 8500, 17000) < 1.0 # 12 kHz cannot ring for seconds in air
+ assert abs(band_rt60(wet_air[:, 0], 707, 1414) - band_rt60(dry_air[:, 0], 707, 1414)) < 0.3 # mids: walls
+ lc = fx.reverb(IMP, dsp.rng(2), rt60=2.0, wet=1.0, dry=0.0, stereo=True, lowcut=200)
+ assert (dsp.butter(lc[:, 0], 'low', 80) ** 2).sum() < 0.01 * (lc[:, 0] ** 2).sum()
+
+
+def test_reverb_tail_does_not_ring():
+ """the long tail must read like a real late field (Gaussian noise), not like a metallic resonator: its long-FFT
+ peakiness stays within 1.5 dB of a pure noise tail, where a pure FDN shows ringing modes"""
+ rt = 3.0
+ ref = ring_stats(dsp.room_ir(dsp.rng(1), rt60=rt, damp=0.0, seconds=rt * 1.6), rt)
+ kw = dict(rt60=rt, damp=0.0, bass=1.0, wet=1.0, dry=0.0, early_db=None, stereo=True, air=0.0)
+ hyb = fx.reverb(IMP, dsp.rng(3), **kw)
+ fdn = fx.reverb(IMP, dsp.rng(3), late='fdn', **kw)
+ assert ring_stats(hyb, rt) < ref + 1.5
+ assert ring_stats(fdn, rt) > ring_stats(hyb, rt) + 1.0 # the measurement does see FDN modes
+ late = hyb[int(0.2 * SR):int(2.0 * SR), 0]
+ fr = late[:len(late) // 480 * 480].reshape(-1, 480)
+ assert abs(np.median((fr ** 4).mean(1) / (fr ** 2).mean(1) ** 2) - 3) < 0.3 # dense: Gaussian, no flutter
+
+
+def test_early_reflections():
+ er = fx.early(IMP, dsp.rng(1), size=0.5)
+ assert er.shape[1] == 2 and np.allclose((er ** 2).sum(axis=0), 1.0, atol=0.05)
+ first = np.nonzero(np.abs(er[:, 0]) > 0.05 * np.abs(er[:, 0]).max())[0][0] / SR
+ assert 0.002 < first < 0.01
+
+
+# ------------------------------------------------------------------------------------------------ fx: dynamics
+def alias_db(y, f0=7000):
+ F = np.abs(np.fft.rfft(y * np.hanning(len(y)))) ** 2
+ f = np.fft.rfftfreq(len(y), 1 / SR)
+ harm = f < 20
+ for k in range(1, 4):
+ harm |= np.abs(f - f0 * k) < 30
+ return 10 * np.log10(F[~harm].sum() / F[harm].sum())
+
+
[email protected]('mode', ['tanh', 'tape', 'hard', 'fold'])
+def test_saturate_oversampled(mode):
+ x = 0.9 * np.sin(2 * np.pi * 7000 * np.arange(SR // 2) / SR)
+ y = fx.saturate(x, 18, mode)
+ assert y.shape == x.shape and np.isfinite(y).all()
+ assert abs(np.sqrt(np.mean(y ** 2)) - np.sqrt(np.mean(x ** 2))) < 1e-6 # level='rms'
+ assert alias_db(y) < -35 and alias_db(y) < alias_db(fx.saturate(x, 18, mode, os=1)) - 20
+ pk = fx.saturate(dsp.sub_drop(0.4), 12, mode, level='peak')
+ assert abs(np.abs(pk).max() - 0.8913) < 1e-3
+ st = fx.saturate(np.stack([x, x], axis=1), 6, mode)
+ assert st.shape == (len(x), 2)
+
+
+def test_saturate_adds_harmonics():
+ x = 0.8 * np.sin(2 * np.pi * 200 * np.arange(SR) / SR)
+ y = fx.saturate(x, 12, 'tanh')
+ F = np.abs(np.fft.rfft(y))
+ assert F[600] > 0.05 * F[200] # odd harmonic (1 Hz bins)
+ y = fx.saturate(x, 12, 'tape')
+ F = np.abs(np.fft.rfft(y))
+ assert F[400] > 0.005 * F[200] # tape bias: even harmonics too
+
+
+def seg_db(y, a, b):
+ return 10 * np.log10(np.mean(y[int(a * SR):int(b * SR)] ** 2) + 1e-20)
+
+
+def test_transient_shaper():
+ r = dsp.rng(3)
+ x = dsp.excite(dsp.modal([300, 800, 1900], [0.08, 0.05, 0.02], [1, 0.6, 0.3], 0.8, r), dsp.contact(0.001, r))
+ y = fx.transient(x, attack_db=6)
+ assert seg_db(y, 0, 0.004) - seg_db(x, 0, 0.004) > 5 and abs(seg_db(y, 0.1, 0.3) - seg_db(x, 0.1, 0.3)) < 1
+ y = fx.transient(x, attack_db=-6)
+ assert seg_db(y, 0, 0.004) - seg_db(x, 0, 0.004) < -4.5
+ y = fx.transient(x, attack_db=0, sustain_db=6)
+ assert abs(seg_db(y, 0, 0.004) - seg_db(x, 0, 0.004)) < 0.5 and seg_db(y, 0.1, 0.3) - seg_db(x, 0.1, 0.3) > 4
+ assert tone.t40_ms(y) > tone.t40_ms(x) + 30 # a longer tail (the v1 impact's weakness)
+ sine = np.sin(2 * np.pi * 440 * np.arange(SR) / SR) * np.minimum(1, np.arange(SR) / 480)
+ y = fx.transient(sine, 6, 6)
+ assert abs(seg_db(y, 0.2, 0.9) - seg_db(sine, 0.2, 0.9)) < 1.0 # a steady tone is left alone
+
+
+def test_limit_ceiling_true_peak():
+ x = dsp.sub_drop(1.0) * 6 + 0.3 * dsp.noise(1.0, dsp.rng(1))
+ y = fx.limit(x, -1.0)
+ assert y.shape == x.shape and np.abs(y).max() <= 10 ** (-1 / 20) + 1e-12
+ up = signal.resample_poly(y, 4, 1)
+ assert 20 * np.log10(np.abs(up).max()) < -0.8
+ q = 0.3 * np.sin(2 * np.pi * 100 * np.arange(SR) / SR)
+ assert np.allclose(fx.limit(q, -1.0), q) # nothing over the ceiling: untouched
+ st = fx.limit(np.stack([x, 0.5 * x], axis=1), -3.0)
+ assert np.abs(st).max() <= 10 ** (-3 / 20) + 1e-12
+ assert np.allclose(st[:, 1], 0.5 * st[:, 0]) # linked: the stereo image does not shift
+
+
+def test_split_sums_flat():
+ bands = fx.split(np.eye(1, 16384, 100)[0], [150, 1200, 6000])
+ assert len(bands) == 4
+ H = np.abs(np.fft.rfft(sum(bands)))
+ assert np.abs(20 * np.log10(H[5:])).max() < 0.05 # an all-pass: flat magnitude
+ lo = np.abs(np.fft.rfft(bands[0]))
+ f = np.fft.rfftfreq(16384, 1 / SR)
+ assert lo[np.argmin(np.abs(f - 2000))] < 0.01 * lo[np.argmin(np.abs(f - 50))]
+
+
+def test_eq_match_moves_tone_toward_target():
+ pk, wh = dsp.noise(1.0, dsp.rng(1), 'pink'), dsp.noise(1.0, dsp.rng(2))
+ d = lambda a, b: np.sqrt(np.mean((tone.profile(a) - tone.profile(b)) ** 2))
+ for ph in ('min', 'linear'):
+ y = fx.eq_match(pk, wh, phase=ph)
+ assert len(y) == len(pk) and np.isfinite(y).all()
+ assert d(y, wh) < 0.25 * d(pk, wh)
+ assert abs(np.sqrt(np.mean(y ** 2)) - np.sqrt(np.mean(pk ** 2))) < 1e-6
+ half = fx.eq_match(pk, wh, amount=0.5)
+ assert d(wh, wh) < d(half, wh) < d(pk, wh)
+ g = fx.match_curve(pk, [wh, dsp.noise(1.0, dsp.rng(3))])
+ assert len(g) == len(tone.CEN) and g[-1] > g[0] + 10 # pink -> white: tilt up
+ assert max(abs(v) for v in fx.match_curve(pk, wh, max_db=6)) <= 6.0
+ hit = dsp.excite(dsp.modal([620, 1490], [0.03, 0.02], [1, .5], 0.2, dsp.rng(4)), dsp.contact(0.001, dsp.rng(4)))
+ y = fx.eq_bands(hit, [0.0] * len(tone.CEN))
+ assert np.max(np.abs(y - hit)) < 1e-3 * np.abs(hit).max() # a flat curve changes nothing
+
+
+# ------------------------------------------------------------------------------------------------ fx: motion
+def test_doppler_pitch_level_pan():
+ src = np.sin(2 * np.pi * 1000 * np.arange(3 * SR) / SR)
+ y = fx.doppler(src, speed=40, dist=3)
+ assert y.ndim == 2 and np.isfinite(y).all()
+ m = y.mean(axis=1)
+ assert abs(peak_hz(m[:8192]) - 1000 * 343 / (343 - 40)) < 12 # approaching: higher
+ assert abs(peak_hz(m[-8192:]) - 1000 * 343 / (343 + 40)) < 12 # receding: lower
+ rms = lambda a: np.sqrt(np.mean(a ** 2))
+ assert rms(y[:8192, 0]) > 3 * rms(y[:8192, 1]) and rms(y[-8192:, 1]) > 3 * rms(y[-8192:, 0])
+ mid = np.argmax(np.abs(m)) / len(m)
+ assert 0.35 < mid < 0.65 # loudest when closest
+ far = fx.doppler(dsp.noise(3, dsp.rng(1)), speed=40, dist=3, stereo=False, air=3.0)
+ near = dsp.butter(far, 'high', 8000)
+ a, b = rms(near[:SR // 4]), rms(near[len(far) // 2 - SR // 8:len(far) // 2 + SR // 8])
+ assert a < 0.3 * b # far away: the highs are absorbed
+
+
+def test_pan_width_decorrelate():
+ x = dsp.noise(1.0, dsp.rng(1))
+ p = fx.pan(x, [(0, -1.0), (1.0, 1.0)])
+ assert p.shape == (len(x), 2)
+ assert np.abs(p[:100, 1]).max() < 0.03 and np.abs(p[-100:, 0]).max() < 0.03
+ assert np.allclose(fx.pan(x, 0.0), np.stack([x, x], axis=1))
+ d = fx.decorrelate(x, 1.0)
+ assert d.shape == (len(x), 2) and abs(np.corrcoef(d[:, 0], d[:, 1])[0, 1]) < 0.7
+ lo = dsp.butter(d, 'low', 100)
+ assert np.corrcoef(lo[:, 0], lo[:, 1])[0, 1] > 0.95 # the low end stays mono
+ w = fx.width(d, [(0, 0.0), (1.0, 2.0)])
+ assert np.allclose(w[:50, 0], w[:50, 1], atol=1e-2)
+ assert np.allclose(fx.width(d, 1.0), d)
+
+
+def test_sweep_filter_moves():
+ y = fx.sweep_filter(dsp.noise(2.0, dsp.rng(1)), 300, 6000, q=8)
+ c = lambda s: np.sum(np.abs(np.fft.rfft(s)) * np.fft.rfftfreq(len(s), 1 / SR)) / np.sum(np.abs(np.fft.rfft(s)))
+ assert c(y[:SR // 4]) < 0.5 * c(y[-SR // 4:])
+ st = fx.sweep_filter(np.ones((1000, 2)), 9000, 200)
+ assert st.shape == (1000, 2)
+
+
+def test_varispeed_and_tape_stop():
+ x = np.sin(2 * np.pi * 1000 * np.arange(SR) / SR)
+ y = fx.varispeed(x, 2.0)
+ assert abs(len(y) - SR / 2) < 3 and abs(peak_hz(y[:8192]) - 2000) < 10
+ y = fx.varispeed(x, [(0, 1.0), (0.5, 0.5)])
+ assert abs(len(y) - SR // 2) < 2 # breakpoints set the output length
+ assert abs(peak_hz(y[:4096]) - 1000 * (1 - 0.5 * 2048 / 24000)) < 15 and abs(peak_hz(y[-4096:]) - 543) < 15
+ t = fx.tape_stop(x, start=0.5, seconds=0.5)
+ assert np.isfinite(t).all() and len(t) > SR * 0.6
+ assert peak_hz(t[int(0.35 * SR):int(0.35 * SR) + 4096]) > 950
+ assert peak_hz(t[-6000:-1500]) < 400 # sliding down to a stop
+ assert abs(t[-1]) < 1e-3
+
+
+# ------------------------------------------------------------------------------------------------ fx: glitch + grains
+def test_crush():
+ y = fx.crush(np.linspace(-1, 1, 1000), bits=3)
+ assert len(np.unique(y)) <= 9
+ s = fx.crush(np.sin(2 * np.pi * 100 * np.arange(4800) / SR), bits=None, rate=8000)
+ assert np.all(s[::6] == s[1::6]) # held for 48000 / 8000 = 6 samples
+ assert np.allclose(fx.crush(s, 4, mix=0.0), s)
+
+
+def test_stutter_and_glitch():
+ x = dsp.fm(300, 1.5, 2, 0.6, decay=0.5)
+ y = fx.stutter(x, 0.1, 0.05, repeats=4, gate=0.8, decay_db=-3)
+ assert len(y) == len(x) + 3 * dsp.n_of(0.05) and np.isfinite(y).all()
+ i0, L = dsp.n_of(0.1), dsp.n_of(0.05)
+ r1, r2 = y[i0:i0 + L], y[i0 + L:i0 + 2 * L]
+ assert np.sqrt(np.mean(r2 ** 2)) < np.sqrt(np.mean(r1 ** 2)) * 0.8
+ assert np.abs(y[i0 + int(0.85 * L):i0 + L]).max() == 0 # gated
+ g = ok(same(lambda r: fx.glitch(x, r, grid=0.03)), np.abs(x).max() + 1e-9)
+ assert len(g) == len(x) and not np.allclose(g, x)
+ st = fx.glitch(np.stack([x, x], axis=1), dsp.rng(1))
+ assert st.shape == (len(x), 2)
+
+
+def test_granular_cloud():
+ src = dsp.bell(880, 1.0, dsp.rng(1), 'glass', decay=0.6)
+ c = same(lambda r: fx.granular(src, 2.0, r, grain=0.05, density=50, pitch=12, pitch_jitter=0.1))
+ assert c.shape == (2 * SR, 2) and np.isfinite(c).all()
+ assert abs(peak_hz(c.mean(axis=1), 1000, 3000) - 1760) < 40 # an octave up
+ assert 0.2 < np.sqrt(np.mean(c ** 2)) / np.sqrt(np.mean(src[:SR // 2] ** 2)) < 3
+ assert abs(np.corrcoef(c[:, 0], c[:, 1])[0, 1]) < 0.95 # spread across the stereo field
+
+
+def test_reverse_swell_lands_on_the_hit():
+ hit = dsp.sub_drop(0.8, dsp.rng(1), click=0.5)
+ y = fx.reverse_swell(hit, dsp.rng(2), seconds=0.9, rt60=1.5, swell_db=-6)
+ assert y.ndim == 2 and np.isfinite(y).all()
+ m = np.abs(y).max(axis=1)
+ env = ndimage.uniform_filter1d(m[:dsp.n_of(0.9)], 960)
+ assert env[-2000:].mean() > 4 * env[:5000].mean() # it swells up into the hit
+ assert abs(dsp.onset(y[dsp.n_of(0.85):], -6) + 0.85 - 0.9) < 0.01 # the hit lands at `seconds`
+ sw = np.abs(y[:dsp.n_of(0.89)]).max()
+ assert abs(20 * np.log10(sw / np.abs(y).max()) + 6) < 1.5
diff --git a/tests/test_forge_integration.py b/tests/test_forge_integration.py
new file mode 100644
index 0000000..c1ae250
--- /dev/null
+++ b/tests/test_forge_integration.py
@@ -0,0 +1,180 @@
+"""forge v2 integration: refsets -> forge refs (held-out guard), critic v2 components in the reports, the CLI wiring,
+and the new dsp / fx tools inside a recipe. Synthetic data, no models (the one model test is marked slow)."""
+import json
+import math
+import os
+import types
+
+import numpy as np
+import pytest
+import soundfile as sf
+
+from earshot.forge import dsp, forge, fx, pick_refs, refsets, report_md, resolve_refs
+from earshot.forge.search import envelope_stats, prescreen_for, render
+
+SR = 48000
+
+
+def _ping(f=1000.0, tau=0.03, dur=0.3, seed=0):
+ t = np.arange(int(dur * SR)) / SR
+ r = np.random.default_rng(seed)
+ return (np.sin(2 * np.pi * f * t) * np.exp(-t / tau) + 1e-3 * r.standard_normal(len(t))) * 0.8
+
+
+def _registry(tmp_path, monkeypatch, cats=('whoosh_fast', 'impact')):
+ """<tmp>/refs/<cat>/{forge,bench}/*.wav -> a registry file EARSHOT_REFSETS points at; returns the registry dict"""
+ root = tmp_path / 'refs'
+ for c in cats:
+ for s, n in (('forge', 3), ('bench', 2)):
+ d = root / c / s
+ d.mkdir(parents=True)
+ for i in range(n):
+ sf.write(str(d / f'{c}_{s}_{i}.wav'), _ping(500 + 100 * i, seed=i), SR)
+ path = str(tmp_path / 'refsets.json')
+ monkeypatch.setenv('EARSHOT_REFSETS', path)
+ refsets.write(refsets.scan(str(root)), path)
+ return refsets.load(path)
+
+
+def test_held_out_flags_registry_bench_files_and_bench_source_kit(tmp_path, monkeypatch):
+ reg = _registry(tmp_path, monkeypatch)
+ fo, be = reg['categories']['impact']['forge'][0], reg['categories']['impact']['bench'][0]
+ kit = str(tmp_path / 'impact_03.wav')
+ assert refsets.held_out([fo, be, kit]) == [be, kit]
+ assert refsets.held_out([fo, be, kit], kit=False) == [be]
+ monkeypatch.setenv('EARSHOT_REFSETS', str(tmp_path / 'none.json')) # no registry: only the kit rule
+ assert refsets.held_out([fo, be, kit]) == [kit]
+
+
+def test_pick_refs_prefers_refset_then_refs_then_recipe_refset_then_recipe_refs(tmp_path, monkeypatch):
+ reg = _registry(tmp_path, monkeypatch)
+ forge_w = reg['categories']['whoosh_fast']['forge']
+ bench_w = reg['categories']['whoosh_fast']['bench']
+ assert pick_refs(refset='whoosh_fast') == (forge_w, 'whoosh_fast')
+ said = []
+ got, cat = pick_refs(refs=forge_w[:1] + bench_w[:1], log=said.append) # explicit refs: bench dropped
+ assert got == forge_w[:1] and cat is None and 'held-out' in said[0]
+ kit_ok, kit_bad = str(tmp_path / 'impact_01.wav'), str(tmp_path / 'impact_03.wav')
+ for p in (kit_ok, kit_bad):
+ sf.write(p, _ping(), SR)
+ rec = types.SimpleNamespace(REFSET='impact', REFS=[kit_ok, kit_bad])
+ assert pick_refs(recipe=rec) == (reg['categories']['impact']['forge'], 'impact')
+ rec.REFSET = 'no_such_category' # unknown category: the kit REFS,
+ assert pick_refs(recipe=rec) == ([kit_ok], None) # minus the bench-source kit file
+ monkeypatch.setenv('EARSHOT_REFSETS', str(tmp_path / 'none.json'))
+ rec.REFSET = 'impact'
+ assert resolve_refs(recipe=rec) == [kit_ok]
+ assert resolve_refs(recipe=types.SimpleNamespace()) is None
+
+
+class V2Critic:
+ """a model-free stand-in that returns rows shaped like critic v2's (v1 terms + env / art / polish components)"""
+ target, alts, tone, tone_target, version, mode = 'a 1 kHz ping', ['a hum'], 0.0, None, 2, 'real'
+ W = {'tone': 1.0, 'env': 1.0, 'art': 1.0, 'polish': 0.75}
+
+ def __init__(self, refs=()):
+ self.refs = list(refs)
+ self.polish_ref = None
+
+ def score(self, xs, exclude=None):
+ rows = []
+ for i, x in enumerate(xs):
+ m = np.asarray(x).mean(1) if np.ndim(x) == 2 else np.asarray(x)
+ f = np.fft.rfftfreq(len(m), 1 / SR)[int(np.argmax(np.abs(np.fft.rfft(m))))]
+ t20 = envelope_stats(x)['t20_ms']
+ env = round(abs(math.log(max(t20, 1.0) / 70.0)), 3)
+ total = 5.0 - 4 * abs(math.log2(max(f, 1.0) / 1000)) - env
+ rows.append({'i': i, 'total': round(total, 3), 'total_v1': round(total + env, 3), 'passes': total > 3.5,
+ 'match_z': 4.0, 'alt_margin': 1.0, 'p_real': 0.9, 'ref_typ_z': None, 'feat_pen': 0.0,
+ 'sounds_like': ['a ping'], 'nearest_ref': None, 'mode': 'real', 'tone_pen': 0.1,
+ 'env_db': 5.0, 'env_ratio': 1.0, 'env_out': 0.0, 'env_pen': env, 'art_pen': 0.0,
+ 'polish_pen': 0.25, 'hard': [], 'polish_words': 'clean', 'summary': f'total {total:.2f}'})
+ return rows
+
+ def score_refs(self, loo=True):
+ return self.score([sf.read(p)[0] for p in self.refs])
+
+ def holdout(self, xs):
+ return [{'audioset_top': [['Ping', 0.5]], 'audioset_synth_max': 0.0} for _ in xs]
+
+
+def _toy():
+ def rend(p, r):
+ t = np.arange(int(0.3 * SR)) / SR
+ return np.sin(2 * np.pi * p['f'] * t) * np.exp(-t / p['tau']) + 1e-4 * r.standard_normal(len(t))
+ return types.SimpleNamespace(__name__='earshot.forge.recipes.toy', SPACE={'f': (300.0, 3000.0, 'log'),
+ 'tau': (0.005, 0.1, 'log')}, DEFAULTS={'f': 600.0, 'tau': 0.05}, render=rend)
+
+
+def test_forge_report_carries_the_v2_components(tmp_path):
+ refs = []
+ for i in range(4):
+ p = str(tmp_path / f'ref_{i}.wav')
+ sf.write(p, _ping(990 + 10 * i, 0.03 + 0.002 * i, seed=i), SR)
+ refs.append(p)
+ c = V2Critic(refs)
+ out = tmp_path / 'made'
+ rep = forge(_toy(), budget=16, pop=8, keep=2, seed=2, out=str(out), log=None, critic=c, prescreen=16)
+ assert rep['critic'] == {'version': 2, 'mode': 'real', 'weights': V2Critic.W} and rep['tone_weight'] == 1.0
+ disk = json.loads((out / 'report.json').read_text(encoding='utf-8'))
+ for k in ('total_v1', 'env_pen', 'art_pen', 'polish_pen', 'hard', 'mode'):
+ assert k in disk['history'][0]['score'] and k in disk['ref_scores'][0], k
+ md = (out / 'report.md').read_text(encoding='utf-8')
+ assert 'critic v2, mode real' in md and '| env pen | art pen | polish pen | hard |' in md
+ assert md.count('| real ref_') == 4
+
+
+def test_cli_passes_refset_mode_and_report_only_tone(monkeypatch, tmp_path):
+ import earshot.forge as F
+ from earshot.cli import main
+ _registry(tmp_path, monkeypatch, cats=('pop',))
+ seen = {}
+
+ def fake(rec, *a, **kw):
+ seen.update(kw, refs=a[2])
+ return {'summary': 'ok', 'picks': []}
+ monkeypatch.setattr(F, 'forge', fake)
+ assert main(['forge', 'pop', '--budget', '8', '--tone', '0', '--mode', 'designed', '--no-bank']) == 0
+ assert seen['refset'] == 'pop' and len(seen['refs']) == 3 # the recipe's REFSET, forge split
+ assert seen['weights'] == {'tone': 0.0} and seen['mode'] == 'designed'
+ assert main(['forge', 'pop', '--budget', '8', '--no-bank']) == 0
+ assert seen['weights'] is None and seen['tone'] == 0.0 # the critic's own v2 weight
+
+
+def test_prescreen_reuses_the_critics_polish_ref():
+ c = types.SimpleNamespace(refs=[], tone_target=None, polish_ref='sentinel')
+ assert prescreen_for(c).pol == 'sentinel'
+
+
+def test_a_recipe_built_on_dsp_and_fx_renders_through_the_search_path():
+ """the round-2 recipes stack dsp sources and fx: a sub drop + air burst through reverb and the limiter"""
+ def rend(p, r):
+ sub = dsp.sub_drop(1.2, r, f0=p['f0'], f1=35.0)
+ hit = dsp.stack([{'x': sub}, {'x': dsp.air(0.4, r), 'gain': -10.0, 't': 0.01}])
+ return fx.limit(fx.reverb(hit, r, rt60=p['rt60'], lowcut=150.0), ceiling_db=-1.0)
+ rec = types.SimpleNamespace(__name__='toy_hit', SPACE={'f0': (60.0, 140.0), 'rt60': (0.5, 2.5)},
+ DEFAULTS={'f0': 90.0, 'rt60': 1.2}, render=rend)
+ x = render(rec, rec.DEFAULTS, 3)
+ assert x is not None and np.all(np.isfinite(x)) and np.abs(x).max() <= 1.0
+ assert len(x) >= int(1.2 * SR)
+
+
+def test_critic_typicality_needs_three_refs(tmp_path):
+ from earshot.forge.critic import Critic
+ refs = []
+ for i in range(2):
+ p = str(tmp_path / f'r{i}.wav')
+ sf.write(p, _ping(800 + 50 * i, 0.04, seed=i), SR)
+ refs.append(p)
+ row = Critic('a short ping', refs=refs).score([_ping(820, 0.04, seed=9)])[0]
+ assert row['ref_typ_z'] is None # 2 refs = one pair, no spread
+
+
+def test_forge_without_refs_still_writes_every_output(tmp_path):
+ """empty refsets (downlifter, glitch) give a critic without refs: designed mode, artifact-only pre-screen"""
+ out = tmp_path / 'made'
+ rep = forge(_toy(), budget=16, pop=8, keep=2, seed=1, out=str(out), log=None, critic=V2Critic(), prescreen=16)
+ for f in ('report.md', 'report.json', 'compare.png', 'audition.wav'):
+ assert (out / f).exists(), f
+ assert rep['ref_scores'] == [] and 'compare_error' not in rep['files']
diff --git a/tests/test_forge_refsets.py b/tests/test_forge_refsets.py
new file mode 100644
index 0000000..a4dc9f3
--- /dev/null
+++ b/tests/test_forge_refsets.py
@@ -0,0 +1,185 @@
+"""Fast tests for earshot.forge.refsets (registry read / scan / write / split checks) and the model-free parts of the
+forge benchmark (eval/forge/bench.py: envelope, artifact and diversity checks). Synthetic data only, no models."""
+import importlib.util
+import json
+import os
+
+import numpy as np
+import pytest
+import soundfile as sf
+
+from earshot.forge import refsets
+
+SR = 48000
+BENCH = os.path.join(os.path.dirname(os.path.dirname(os.path.abspath(__file__))), 'eval', 'forge', 'bench.py')
+
+
+def _wav(path, x):
+ os.makedirs(os.path.dirname(path), exist_ok=True)
+ sf.write(path, x, SR, subtype='PCM_24')
+ return path
+
+
+def _ping(f=1000.0, tau=0.05, n=0.4, pre=0.01, seed=0):
+ t = np.arange(int(n * SR)) / SR
+ x = np.sin(2 * np.pi * f * t) * np.exp(-t / tau) + 1e-4 * np.random.default_rng(seed).standard_normal(len(t))
+ x = np.concatenate([np.zeros(int(pre * SR)), x])
+ k = int(0.002 * SR)
+ x[int(pre * SR):int(pre * SR) + k] *= np.linspace(0, 1, k)
+ x[-int(0.01 * SR):] *= np.linspace(1, 0, int(0.01 * SR))
+ return 0.8 * x / np.abs(x).max()
+
+
+def layout(tmp_path):
+ root = tmp_path / 'refs'
+ for i in range(3):
+ _wav(str(root / 'pop' / 'forge' / f'f{i}.wav'), _ping(800 + 100 * i))
+ _wav(str(root / 'pop' / 'bench' / f'b{i}.wav'), _ping(900 + 100 * i, seed=i + 5))
+ _wav(str(root / 'click' / 'bench' / 'c0.wav'), _ping(3000, 0.01))
+ (root / '_work').mkdir()
+ return root
+
+
+def test_missing_registry_is_clear(tmp_path, monkeypatch):
+ monkeypatch.setenv('EARSHOT_REFSETS', str(tmp_path / 'none.json'))
+ with pytest.raises(refsets.RefsetsMissing, match='bring'):
+ refsets.refset('pop')
+ with pytest.raises(FileNotFoundError): # RefsetsMissing is a FileNotFoundError
+ refsets.categories()
+
+
+def test_scan_write_and_read(layout, tmp_path, monkeypatch):
+ reg = refsets.scan(str(layout))
+ assert sorted(reg['categories']) == ['click', 'pop'] # _work is skipped
+ assert reg['categories']['click']['forge'] == []
+ path = refsets.write(reg, str(tmp_path / 'reg.json'))
+ monkeypatch.setenv('EARSHOT_REFSETS', path)
+ assert refsets.registry_path() == os.path.abspath(path)
+ assert refsets.categories() == ['click', 'pop']
+ assert [os.path.basename(p) for p in refsets.refset('pop')] == ['f0.wav', 'f1.wav', 'f2.wav']
+ assert [os.path.basename(p) for p in refsets.refset('pop', split='bench')] == ['b0.wav', 'b1.wav', 'b2.wav']
+ assert refsets.refset('click') == [] and refsets.note('click') == ''
+ with pytest.raises(KeyError, match='known: click, pop'):
+ refsets.refset('whoosh_fast')
+ with pytest.raises(ValueError):
+ refsets.refset('pop', split='test')
+
+
+def test_relative_paths_and_missing_files(tmp_path, monkeypatch):
+ _wav(str(tmp_path / 'a' / 'x.wav'), _ping())
+ reg = {'version': 1, 'categories': {'pop': {'forge': ['a/x.wav'], 'bench': ['a/gone.wav'], 'note': 'thin'}}}
+ path = tmp_path / 'reg.json'
+ path.write_text(json.dumps(reg))
+ monkeypatch.setenv('EARSHOT_REFSETS', str(path))
+ assert refsets.refset('pop')[0].endswith('a/x.wav') and os.path.isabs(refsets.refset('pop')[0])
+ assert refsets.note('pop') == 'thin'
+ with pytest.raises(refsets.RefsetsMissing, match='missing'):
+ refsets.refset('pop', split='bench')
+
+
+def test_split_checks(tmp_path):
+ reg = {'categories': {'pop': {'forge': ['a.wav'], 'bench': ['a.wav', 'b.wav']},
+ 'tap': {'forge': ['c.wav'], 'bench': ['d.wav']}}}
+ assert refsets.check_split(reg) == ['pop: a.wav in both splits']
+ with pytest.raises(ValueError, match='leaks'):
+ refsets.write(reg, str(tmp_path / 'r.json'))
+ reg['categories']['pop']['bench'] = ['b.wav']
+ assert refsets.check_split(reg) == []
+ fam = {'a.wav': 'author1', 'b.wav': 'author2', 'c.wav': 'author3', 'd.wav': 'author1'} # a sister take of a.wav
+ bad = refsets.check_split(reg, fam)
+ assert len(bad) == 1 and 'author1' in bad[0] and "['pop']" in bad[0] and "['tap']" in bad[0]
+
+
+def test_categories_cover_the_launch_set():
+ for c in ('whoosh_fast', 'whoosh_big', 'swish', 'swipe', 'riser', 'downlifter', 'impact', 'sub_drop', 'click', 'tap',
+ 'toggle', 'hover_tick', 'pop', 'notify', 'shimmer', 'typing_key', 'glitch', 'logo_sting'):
+ assert c in refsets.CATEGORIES
+
+
+# ------------------------------------------------------------------------------- bench.py model-free checks
[email protected](scope='module')
+def bench():
+ spec = importlib.util.spec_from_file_location('forge_bench', BENCH)
+ m = importlib.util.module_from_spec(spec)
+ spec.loader.exec_module(m)
+ return m
+
+
+def test_envelope_features(bench):
+ x = _ping(1000, 0.05, pre=0.1)
+ f = bench.env_feats(x)
+ assert abs(f['onset_s'] - 0.1) < 0.005
+ assert f['attack_ms'] <= 4
+ assert 200 < f['t40_ms'] < 260 # 40 dB of an exp(-t / 50 ms) decay = 50 ms * ln(100) = 230 ms
+ slow = bench.env_feats(_ping(1000, 0.2, n=1.2))
+ assert slow['t40_ms'] > 3 * f['t40_ms'] and slow['active_s'] > f['active_s']
+
+
+def test_env_distance_sees_a_short_tail(bench):
+ refs = [_ping(1000, tau, seed=i) for i, tau in enumerate((0.08, 0.1, 0.12))]
+ er = bench.EnvRef(refs)
+ same = er.check(_ping(1000, 0.1, seed=9))
+ short = er.check(_ping(1000, 0.02, seed=9))
+ assert same['env_oct'] < 0.3 and short['env_oct'] > 1.0
+ assert short['curve_db'] > same['curve_db'] + 5
+ assert short['t40_ratio'] < 0.35
+ loo = er.check(refs[1], exclude=1)
+ assert loo['env_oct'] < 0.5
+
+
+def test_artifacts(bench):
+ clean = _ping()
+ assert bench.artifacts(clean)['flags'] == []
+ hard = np.sign(np.cos(2 * np.pi * 500 * np.arange(len(clean)) / SR)) # clipped square from a full-scale sample
+ fl = bench.artifacts(hard)['flags']
+ assert 'clipped' in fl and 'start click' in fl
+ cut = _ping(1000, 0.5)[:int(0.15 * SR)] # ends while still ringing
+ assert 'tail cut' in bench.artifacts(cut)['flags'] and 'end click' in bench.artifacts(cut)['flags']
+ dc = clean + 0.2
+ assert 'dc offset' in bench.artifacts(dc)['flags']
+ late = np.concatenate([np.zeros(int(0.3 * SR)), clean])
+ assert 'late start' in bench.artifacts(late)['flags']
+ t = np.arange(int(1.0 * SR)) / SR # a swell from the first sample: not late
+ swell = 0.8 * np.sin(2 * np.pi * 300 * t) * np.minimum(1, t / 0.8) ** 1.5
+ swell[-480:] *= np.linspace(1, 0, 480)
+ assert 'late start' not in bench.artifacts(swell)['flags']
+ assert 'tail cut' in bench.artifacts(cut)['flags'] and 'tail cut' not in bench.artifacts(cut, 'riser')['flags']
+
+
+def test_diversity(bench):
+ xs = [_ping(1000), _ping(1000, seed=1), _ping(3000, 0.01)]
+ E = np.eye(3)
+ E[1] = E[0] # two identical embeddings
+ d = bench.diversity(xs, E)
+ assert d['n'] == 3 and d['near_duplicates'] == [[0, 1]]
+ assert d['cos_max'] == pytest.approx(1.0)
+ assert bench.diversity(xs[:1], E[:1])['n'] == 1
+
+
+def test_summary_inside_real(bench):
+ real = [{'tone_db': 4.0, 'typ_z': -0.5}, {'tone_db': 6.0, 'typ_z': -1.5}]
+ made = [{'tone_db': 5.0, 'typ_z': -1.0}, {'tone_db': 12.0, 'typ_z': -4.0}]
+ s = bench.inside_real(made, real, metrics=(('tone_db', 'low'), ('typ_z', 'high')))
+ assert s['per_sound'] == [2, 0] and s['n_checks'] == 2
+
+
+def test_inside_real_quantile(bench):
+ real = [{'tone_db': v} for v in (2.0, 3.0, 4.0, 5.0, 20.0)] # one far outlier among the real rows
+ made = [{'tone_db': 10.0}]
+ assert bench.inside_real(made, real, metrics=(('tone_db', 'low'),), q=1.0)['per_sound'] == [1]
+ assert bench.inside_real(made, real, metrics=(('tone_db', 'low'),), q=0.8)['per_sound'] == [0]
+
+
+def test_texture(bench):
+ rng = np.random.default_rng(3)
+ dry = _ping(2000, 0.1, n=0.8) # digital silence before the attack
+ tx = bench.texture(dry)
+ assert tx['stereo_corr'] == 1.0 and tx['floor_db'] < -60
+ floor = dry + 10 ** (-50 / 20) * rng.standard_normal(len(dry))
+ assert -60 < bench.texture(floor)['floor_db'] < -40
+ st = np.stack([floor, floor + 0.05 * rng.standard_normal(len(floor))], 1)
+ assert bench.texture(st)['stereo_corr'] < 0.99
+ t = np.arange(int(0.8 * SR)) / SR # bright partial dies first: darkens
+ dark = np.sin(2 * np.pi * 500 * t) * np.exp(-t / 0.15) + np.sin(2 * np.pi * 6000 * t) * np.exp(-t / 0.02)
+ assert bench.texture(dark)['darken'] < -1
diff --git a/tests/test_forge_search.py b/tests/test_forge_search.py
new file mode 100644
index 0000000..72afe03
--- /dev/null
+++ b/tests/test_forge_search.py
@@ -0,0 +1,276 @@
+"""Fast tests for forge search v2: CMA-ES with restarts, the model-free pre-screen, threaded rendering, envelope and
+artifact measurements, diverse picks by sound, the seed bank, forge() outputs (report.md, compare.png, audition.wav),
+render without search and the CLI. Toy recipes and fake critics only: no models are loaded."""
+import json
+import math
+import os
+import types
+
+import numpy as np
+import soundfile as sf
+
+from earshot.forge import audition, forge, render_params, report_md
+from earshot.forge import search
+from earshot.forge.search import (Prescreen, artifacts, describe, diverse, envelope_stats, load_seeds, optimize,
+ render, render_many, save_seeds, sound_dist)
+
+SR = 48000
+
+
+def _ping(f, tau, n=None, fade=True, r=None):
+ n = n or int(0.25 * SR)
+ t = np.arange(n) / SR
+ x = np.sin(2 * np.pi * f * t) * np.exp(-t / tau) * np.minimum(1.0, t / 0.001)
+ if r is not None:
+ x = x + 0.0005 * r.standard_normal(n)
+ if fade:
+ x[-480:] *= np.linspace(1, 0, 480)
+ return x
+
+
+def _toy_render(p, r):
+ return _ping(p['f'], p['tau'], r=r) * {'soft': 0.2, 'hard': 1.0}[p['shape']]
+
+
+def toy(**over):
+ m = types.SimpleNamespace(__name__='earshot.forge.recipes.toy',
+ SPACE={'f': (200.0, 4000.0, 'log'), 'tau': (0.005, 0.2, 'log'), 'shape': ['soft', 'hard'],
+ 'n': (1, 9, 'int')},
+ DEFAULTS={'f': 300.0, 'tau': 0.15, 'shape': 'soft', 'n': 3}, render=_toy_render)
+ for k, v in over.items():
+ setattr(m, k, v)
+ return m
+
+
+def _peak_hz(x):
+ m = np.asarray(x).mean(axis=1) if np.ndim(x) == 2 else np.asarray(x)
+ return np.fft.rfftfreq(len(m), 1 / SR)[int(np.argmax(np.abs(np.fft.rfft(m))))]
+
+
+class FakeCritic:
+ """scores by closeness to 1 kHz and a 30 ms decay (no models); counts how many sounds it heard"""
+ target, alts, refs, tone, tone_target = 'a 1 kHz ping', ['a low hum'], [], 0.0, None
+
+ def __init__(self):
+ self.heard = 0
+
+ def score(self, xs, exclude=None):
+ rows = []
+ for i, x in enumerate(xs):
+ self.heard += 1
+ tau = max(1e-3, envelope_stats(x)['t20_ms'] / 1000 / 2.303)
+ total = 5.0 - 4 * abs(math.log2(max(_peak_hz(x), 1.0) / 1000)) - abs(math.log(tau / 0.03))
+ rows.append({'i': i, 'total': round(total, 3), 'passes': total > 3.5, 'match_z': total, 'alt_margin': 1.0,
+ 'p_real': 0.9, 'ref_typ_z': None, 'feat_pen': 0.0, 'sounds_like': ['a ping'],
+ 'nearest_ref': None, 'summary': f'total {total:.2f}'})
+ return rows
+
+ def score_refs(self, loo=True):
+ return self.score([sf.read(p)[0] for p in self.refs])
+
+ def holdout(self, xs):
+ return [{'audioset_top': [['Ping', 0.5]], 'audioset_synth_max': 0.0} for _ in xs]
+
+
+def _refs(tmp_path, n=4):
+ """four real-like 1 kHz pings (slightly different decays and pitches) written as WAVs"""
+ out = []
+ for i in range(n):
+ p = str(tmp_path / f'ref_{i}.wav')
+ sf.write(p, _ping(980 + 15 * i, 0.028 + 0.003 * i, r=np.random.default_rng(i)) * 0.8, SR)
+ out.append(p)
+ return out
+
+
+# ---------------------------------------------------------------------------------------------- CMA-ES
+def test_cma_improves_on_the_default_and_is_deterministic():
+ rec = toy()
+ base = FakeCritic().score([render(rec, rec.DEFAULTS, 0)])[0]['total']
+ c = FakeCritic()
+ a = optimize(rec, c, budget=96, pop=16, seed=3, strategy='cma')
+ b = optimize(rec, FakeCritic(), budget=96, pop=16, seed=3, strategy='cma')
+ assert len(a['history']) == 96 and a['generations'] == 6 and c.heard == 96
+ assert a['best'][0]['score']['total'] > base + 2
+ assert [h['params'] for h in a['history']] == [h['params'] for h in b['history']]
+ assert a['history'][0]['params'] == {k: rec.DEFAULTS[k] for k in rec.SPACE} # the default is always scored
+ assert all('sigma' in g for g in a['gens']) and a['strategy'] == 'cma'
+
+
+def test_cma_restarts_when_stuck():
+ res = optimize(toy(), FakeCritic(), budget=128, pop=8, seed=1, strategy='cma', patience=1)
+ assert res['restarts'] >= 1 and any(g.get('restart') for g in res['gens'])
+ for h in res['history']:
+ assert 200.0 <= h['params']['f'] <= 4000.0 and h['params']['shape'] in ('soft', 'hard')
+
+
+def test_mirror_keeps_points_in_the_unit_cube():
+ x = np.array([-0.3, 1.4, 0.5, 2.6, -1.2])
+ u = search._mirror(x)
+ assert np.allclose(u, [0.3, 0.6, 0.5, 0.6, 0.8]) and (u >= 0).all() and (u <= 1).all()
+
+
+# ---------------------------------------------------------------------------------------------- pre-screen
+def test_prescreen_scores_only_the_best_pop_and_keeps_the_starts(tmp_path):
+ c = FakeCritic()
+ pre = Prescreen(_refs(tmp_path))
+ res = optimize(toy(), c, budget=48, pop=8, seed=2, strategy='cma', prescreen=32, pre=pre)
+ assert c.heard == 48 and len(res['history']) == 48
+ assert res['timing']['rendered'] == 6 * 32 and res['prescreen'] == 32
+ assert res['history'][0]['params']['f'] == 300.0 # DEFAULTS scored in gen 0
+ assert all('pre' in h and 'desc' in h for h in res['history'])
+ late = [h for h in res['history'] if h['gen'] >= 3]
+ assert np.median([h['pre']['tone_db'] for h in late]) < np.median([h['pre']['tone_db'] for h in res['history'][:8]])
+
+
+def test_prescreen_ranks_a_ref_like_sound_first(tmp_path):
+ pre = Prescreen(_refs(tmp_path))
+ like, other, broken = _ping(1000, 0.03), _ping(250, 0.15), _ping(1000, 0.03, fade=False)
+ broken[:20] = 0.9
+ rows = pre.score([like, other, broken, None])
+ assert rows[0]['pre'] > rows[1]['pre'] and rows[0]['pre'] > rows[2]['pre'] and rows[3]['pre'] == -np.inf
+ assert rows[0]['tone_db'] < rows[1]['tone_db']
+ assert 'start_jump' in rows[2]['art']['flags']
+ assert pre.nearest(rows[0]['desc']) is not None
+ assert Prescreen([], polish=False).score([like])[0]['pre'] == 0.0 # no refs: artifacts only
+ nr = Prescreen([]).score([like, broken]) # polish's absolute limits
+ assert nr[0]['pre'] <= 0 and 'tone_db' not in nr[0]
+
+
+def test_prescreen_uses_the_polish_verdict_when_there_is_one(tmp_path):
+ class Pol:
+ def check(self, x):
+ return {'art_pen': 0.5, 'polish_pen': 1.0, 'hard': ['clicks']}
+ pre = Prescreen(_refs(tmp_path), polish_ref=Pol())
+ r = pre.row(_ping(1000, 0.03))
+ assert r['polish']['hard'] == ['clicks']
+ base = Prescreen(_refs(tmp_path), polish=False).row(_ping(1000, 0.03))
+ assert abs((base['pre'] - r['pre']) - (0.5 + 0.75 + 1.0)) < 1e-6
+
+
+def test_render_many_matches_serial_rendering():
+ rec = toy()
+ jobs = [({'f': 300.0 + 100 * i, 'tau': 0.03, 'shape': 'hard', 'n': 1}, 11 + i) for i in range(6)]
+ a = render_many(rec, jobs, workers=1)
+ b = render_many(rec, jobs, workers=3)
+ assert all(np.array_equal(x, y) for x, y in zip(a, b))
+
+
+def test_es_with_threads_is_unchanged():
+ a = optimize(toy(), FakeCritic(), budget=32, pop=16, seed=5, workers=1)
+ b = optimize(toy(), FakeCritic(), budget=32, pop=16, seed=5, workers=4)
+ assert [h['params'] for h in a['history']] == [h['params'] for h in b['history']]
+
+
+# ---------------------------------------------------------------------------------------------- measurements
+def test_envelope_stats_of_an_exponential_decay():
+ tau = 0.04
+ e = envelope_stats(_ping(1000, tau, n=SR))
+ assert abs(e['t20_ms'] - 2.303 * tau * 1000) < 8 and abs(e['t40_ms'] - 4.605 * tau * 1000) < 12
+ assert e['attack_ms'] <= 4 and e['t10_ms'] < e['t20_ms'] < e['t40_ms']
+
+
+def test_artifacts_flag_clicks_cuts_gaps_and_clipping():
+ clean = _ping(1000, 0.03)
+ assert artifacts(clean)['flags'] == {} and artifacts(clean)['penalty'] == 0
+ cut = _ping(1000, 0.3, n=SR // 10, fade=False)
+ assert 'cut_end' in artifacts(cut)['flags']
+ gap = clean.copy()
+ gap[2400:3000] = 0.0
+ assert 'gap' in artifacts(gap)['flags']
+ clip = np.clip(3 * clean, -1, 1)
+ assert 'flat_top' in artifacts(clip)['flags']
+ dc = clean + 0.2
+ assert 'dc' in artifacts(dc)['flags']
+
+
+def test_sound_dist_separates_different_sounds():
+ a, b, c = describe(_ping(1000, 0.03)), describe(_ping(1010, 0.031)), describe(_ping(300, 0.15))
+ assert sound_dist(a, b) < 2.0 < sound_dist(a, c)
+
+
+def test_diverse_prefers_different_sounds_then_fills_by_params():
+ def h(u, total, x):
+ return {'u': np.array(u), 'score': {'total': total, 'passes': True}, 'desc': describe(x)}
+ same = _ping(1000, 0.03)
+ res = [h([0.1, 0.1], 5.0, same), h([0.9, 0.9], 4.9, same), h([0.5, 0.2], 4.0, _ping(300, 0.15))]
+ picks = diverse(res, toy(), k=2)
+ assert [p['score']['total'] for p in picks] == [5.0, 4.0] # the sound twin is skipped
+ picks3 = diverse(res, toy(), k=3)
+ assert len(picks3) == 3 # filled by the v1 rule
+
+
+# ---------------------------------------------------------------------------------------------- seed bank
+def test_seed_bank_round_trip_and_stays_small(tmp_path):
+ folder = str(tmp_path)
+ picks = [{'params': {'f': 1000.0 + i, 'tau': 0.03}, 'score': {'total': 3.0 + i, 'passes': True}}
+ for i in range(8)]
+ save_seeds('toy', picks, 'a ping', ['r1.wav'], folder=folder, per_context=3)
+ bank = json.loads((tmp_path / 'toy.json').read_text(encoding='utf-8'))
+ assert len(bank['entries']) == 3 and bank['entries'][0]['params']['f'] == 1007.0
+ save_seeds('toy', [{'params': {'f': 500.0}, 'score': {'total': 9.0}, 'robust': {'mean': 1.0, 'pass_rate': 0}}],
+ 'another target', [], folder=folder, per_context=3)
+ got = load_seeds('toy', 'a ping', ['r1.wav'], folder=folder)
+ assert got[0]['f'] == 1007.0 and {'f': 500.0} in got # same context first
+ got2 = load_seeds('toy', 'another target', [], folder=folder)
+ assert got2[0] == {'f': 500.0}
+ assert load_seeds('nothing_here', folder=folder) == []
+
+
+# ---------------------------------------------------------------------------------------------- forge outputs
+def test_forge_writes_report_md_compare_png_and_audition(tmp_path):
+ c = FakeCritic()
+ c.refs = _refs(tmp_path)
+ out = tmp_path / 'made'
+ rep = forge(toy(), budget=32, pop=8, keep=2, seed=4, out=str(out), log=None, critic=c, robust=2, shortlist=2,
+ strategy='cma', prescreen=24)
+ assert len(rep['picks']) == 2 and rep['strategy'] == 'cma' and rep['prescreen'] == 24
+ for f in ('report.md', 'report.json', 'compare.png', 'audition.wav', 'audition.txt'):
+ assert (out / f).exists(), f
+ md = (out / 'report.md').read_text(encoding='utf-8')
+ assert '## Search' in md and '## Picks' in md and '## Gap to the real references' in md
+ assert md.count('| real ref_') == 4 and 'measured vs the refs' in md
+ cues = (out / 'audition.txt').read_text(encoding='utf-8').splitlines()[1:]
+ kinds = [ln.split()[2] for ln in cues]
+ assert kinds == ['made', 'ref', 'made', 'ref']
+ x, sr = sf.read(str(out / 'audition.wav'))
+ assert sr == SR and x.shape[1] == 2 and abs(20 * np.log10(np.abs(x).max()) + 3) < 0.1
+ for p in rep['picks']:
+ assert p['robust']['n'] == 2 and p['measured']['nearest_ref'].startswith('ref_')
+ assert rep['gap'].startswith('Measured gap')
+ disk = json.loads((out / 'report.json').read_text(encoding='utf-8'))
+ assert len(disk['history']) == 32 and disk['timing']['scored'] == 32 and len(disk['gens']) == 4
+
+
+def test_report_md_without_refs():
+ rep = forge(toy(), budget=16, pop=8, keep=1, seed=1, log=None, critic=FakeCritic())
+ md = report_md(rep)
+ assert md.startswith("# forge toy: 'a 1 kHz ping'") and '## Gap' not in md
+
+
+def test_audition_alternates_and_peak_matches(tmp_path):
+ made = [('m1', 0.1 * _ping(1000, 0.03)), ('m2', _ping(800, 0.02))]
+ refs = [('r1', 0.5 * _ping(990, 0.03))]
+ a = audition(str(tmp_path / 'a.wav'), made, refs, [0, None])
+ assert [c['kind'] for c in a['cues']] == ['made', 'ref', 'made']
+ assert a['cues'][1]['t'] > a['cues'][0]['t']
+
+
+def test_render_params_from_a_report_and_from_a_dict(tmp_path):
+ rep = {'picks': [{'params': {'f': 1000.0, 'tau': 0.03, 'shape': 'hard', 'n': 1}, 'seed': 42}]}
+ r = render_params(toy(), rep, out=str(tmp_path))
+ assert r['files'][0]['seed'] == 42 and os.path.exists(r['files'][0]['file'])
+ r2 = render_params(toy(), {'f': 500.0}, seeds=[1, 2], out=str(tmp_path), refs=_refs(tmp_path))
+ assert [f['seed'] for f in r2['files']] == [1, 2] and 'tone_db' in r2['files'][0]['measured']
+ a = sf.read(r2['files'][0]['file'])[0]
+ assert abs(_peak_hz(a) - 500) < 10 # DEFAULTS filled the rest
+
+
+def test_cli_list_and_render(tmp_path, capsys):
+ from earshot.cli import main
+ assert main(['forge', '--list']) == 0
+ out = capsys.readouterr().out
+ assert 'keypress: ' in out and 'recipes (name: TARGET)' in out
+ pj = tmp_path / 'p.json'
+ pj.write_text(json.dumps({}), encoding='utf-8')
+ assert main(['forge', '--render', 'pop', '--params', str(pj), '--seeds', '3', '--dir', str(tmp_path)]) == 0
+ assert (tmp_path / 'pop_r01_s3.wav').exists()