Skip to content

Generate clean 2D game sprites & animation atlases — component-row pipeline: state rows, alpha cleanup, frame extraction, runtime atlases. Codex/Claude skill.

fix: preserve exact AAC container frame duration

Soohong Kimddec03cb2fa9
diff --git a/sprite_gen/audio/processing.py b/sprite_gen/audio/processing.py
index 093e186..e701d40 100644
--- a/sprite_gen/audio/processing.py
+++ b/sprite_gen/audio/processing.py
@@ -15,10 +15,11 @@ from sprite_gen._deps import np
from sprite_gen.audio import playback
RATE = 48000
+PROCESSING_VERSION = 3
def require_tools():
- for name in ('ffmpeg',):
+ for name in ('ffmpeg', 'ffprobe'):
if not shutil.which(name):
raise ValueError(f'{name} is required on PATH for sound processing')
encoders = _ffmpeg(['-encoders']).stdout.decode(errors='replace')
@@ -49,6 +50,19 @@ def _pcm_args():
return ['-f', 'f32le', '-ar', str(RATE), '-ac', '2', '-i', 'pipe:0']
+def _check_m4a_duration(path, frames):
+ result = subprocess.run(['ffprobe', '-v', 'error', '-select_streams', 'a:0',
+ '-show_entries', 'stream=duration_ts,time_base', '-of', 'json', str(path)],
+ capture_output=True, timeout=180)
+ try:
+ stream, = json.loads(result.stdout)['streams']
+ except (ValueError, KeyError, TypeError):
+ stream = None
+ if (result.returncode or not isinstance(stream, dict) or
+ (stream.get('time_base'), stream.get('duration_ts')) != (f'1/{RATE}', frames)):
+ raise ValueError(f'm4a container duration must match the declared {frames} frames at {RATE} Hz')
+
+
def measure(samples, target=-18):
result = _ffmpeg([*_pcm_args(), '-af', f'loudnorm=I={target}:TP=-2:LRA=11:print_format=json',
'-f', 'null', '-'], samples.astype('<f4').tobytes())
@@ -105,11 +119,15 @@ def export(source: Path, out_dir: Path, *, target_lufs, loop, crossfade):
'overlap_frames': round(crossfade*RATE) if loop else 0}, loop)
samples, corrections = normalize(samples, target_lufs, loop)
files = {}
- for fmt, codec in (('ogg', ['-c:a', 'libopus', '-b:a', '160k']), ('m4a', ['-c:a', 'aac', '-b:a', '192k', '-movflags', '+faststart'])):
+ for fmt, codec in (('ogg', ['-c:a', 'libopus', '-b:a', '160k']),
+ ('m4a', ['-c:a', 'aac', '-b:a', '192k', '-movflags', '+faststart',
+ '-movie_timescale', str(RATE)])):
path = out_dir / f'sound.{fmt}'
encoded_samples, ceiling = samples, -2.0
for attempt in range(3):
_ffmpeg([*_pcm_args(), '-map_metadata', '-1', *codec, str(path)], encoded_samples.astype('<f4').tobytes())
+ if fmt == 'm4a':
+ _check_m4a_duration(path, frames['end_frame']-frames['start_frame'])
decoded = decode(path)
if len(decoded) < len(samples):
raise ValueError(f'{fmt} truncated the audio')
diff --git a/sprite_gen/audio/sound.py b/sprite_gen/audio/sound.py
index da07b63..2681b7a 100644
--- a/sprite_gen/audio/sound.py
+++ b/sprite_gen/audio/sound.py
@@ -27,7 +27,7 @@ def generate(*, kind, prompt, duration, event, out_dir, cache_dir=None, loop=Non
raise ValueError('crossfade must be positive and shorter than half the duration')
out_dir = Path(out_dir).expanduser().resolve()
recipe = {'request': cache.digest(request), 'event': event, 'loop': loop,
- 'target_lufs': target, 'crossfade_seconds': crossfade, 'processing_version': playback.VERSION}
+ 'target_lufs': target, 'crossfade_seconds': crossfade, 'processing_version': processing.PROCESSING_VERSION}
if out_dir.exists():
manifest = json.loads((out_dir / 'sound.json').read_text()) if (out_dir / 'sound.json').is_file() else {}
if manifest.get('recipe') != recipe:
diff --git a/tests/audio/test_sound.py b/tests/audio/test_sound.py
index 62e5d16..e225197 100644
--- a/tests/audio/test_sound.py
+++ b/tests/audio/test_sound.py
@@ -3,6 +3,7 @@ from concurrent.futures import ThreadPoolExecutor
import io
import http.client
import json
+import subprocess
import threading
import urllib.error
import wave
@@ -230,14 +231,50 @@ def test_export_reserves_overlap_for_minimum_playback_rate(tmp_path, monkeypatch
processing.export(tmp_path/'source', tmp_path, target_lufs=-20, loop=True, crossfade=overlap/48000)
-def test_old_processed_output_is_rejected_but_raw_cache_survives(tmp_path, monkeypatch):
[email protected]('millisecond_timescale', [False, True])
+def test_m4a_container_preserves_declared_frames(tmp_path, monkeypatch, millisecond_timescale):
+ frames = 144003
+ source = tmp_path/'source.wav'
+ source.write_bytes(wav_bytes(frames/48000))
+ if millisecond_timescale:
+ encode = processing._ffmpeg
+ def old_timescale(args, data=None):
+ if str(args[-1]).endswith('.m4a'):
+ args = [*args[:-1], '-movie_timescale', '1000', args[-1]]
+ return encode(args, data)
+ monkeypatch.setattr(processing, '_ffmpeg', old_timescale)
+ with pytest.raises(ValueError, match='m4a container duration'):
+ processing.export(source, tmp_path, target_lufs=-20, loop=True, crossfade=.25)
+ return
+ result = processing.export(source, tmp_path, target_lufs=-20, loop=True, crossfade=.25)
+ probe = subprocess.run(['ffprobe', '-v', 'error', '-select_streams', 'a:0',
+ '-show_entries', 'stream=duration_ts,time_base', '-of', 'json',
+ str(tmp_path/result['files']['m4a']['path'])],
+ capture_output=True, check=True)
+ stream = json.loads(probe.stdout)['streams'][0]
+ assert stream == {'time_base': '1/48000', 'duration_ts': frames}
+ assert result['playback']['end_frame'] == frames
+
+
+def test_missing_ffprobe_preflights_before_generation(tmp_path, monkeypatch):
+ which = processing.shutil.which
+ monkeypatch.setattr(processing.shutil, 'which', lambda name: None if name == 'ffprobe' else which(name))
+ monkeypatch.setattr(provider, 'generate', lambda _: pytest.fail('generation before tool preflight'))
+ with pytest.raises(ValueError, match='ffprobe is required'):
+ sound.generate(kind='sfx', prompt='synthetic', duration=3, event='hit',
+ cache_dir=tmp_path/'cache', out_dir=tmp_path/'out')
+
+
[email protected]('old_version', [1, 2])
+def test_old_processed_output_is_rejected_but_raw_cache_survives(tmp_path, monkeypatch, old_version):
calls = []
monkeypatch.setattr(provider, 'generate', lambda req: (calls.append(req) or wav_bytes(), {}))
kw = dict(kind='bgm', prompt='synthetic', duration=3, event='music',
cache_dir=tmp_path/'cache', out_dir=tmp_path/'out')
result = sound.generate(**kw)
- assert result['version'] == result['recipe']['processing_version'] == 2
- old = {**result, 'version': 1, 'recipe': {**result['recipe'], 'processing_version': 1}}
+ assert result['version'] == 2
+ assert result['recipe']['processing_version'] == 3
+ old = {**result, 'version': old_version, 'recipe': {**result['recipe'], 'processing_version': old_version}}
(kw['out_dir']/'sound.json').write_text(json.dumps(old))
with pytest.raises(ValueError, match='different sound'):
sound.generate(**kw)