Skip to content

Speakr is a personal, self-hosted web application designed for transcribing audio recordings

Python

Converge upload_file onto the shared transcription resolver and unify empty-language handling to fall through tag/folder/account defaults instead of forcing auto-detect

Murtaza Nasirf37469d6b586
diff --git a/src/api/recordings.py b/src/api/recordings.py
index e297ba1f..ac83e23d 100644
--- a/src/api/recordings.py
+++ b/src/api/recordings.py
@@ -2697,61 +2697,27 @@ def upload_file():
except (ValueError, TypeError):
max_speakers = None
- # Apply precedence hierarchy: user input > tag defaults > folder defaults > environment variables > user defaults > auto-detect
-
- # Apply folder defaults first (lower priority than tags)
- if selected_folder and not selected_tags:
- # Only apply folder defaults if no tags are selected (tags take priority)
- if not language and selected_folder.default_language:
- language = selected_folder.default_language
- if min_speakers is None and selected_folder.default_min_speakers:
- min_speakers = selected_folder.default_min_speakers
- if max_speakers is None and selected_folder.default_max_speakers:
- max_speakers = selected_folder.default_max_speakers
- if not hotwords and selected_folder.default_hotwords:
- hotwords = selected_folder.default_hotwords
- if not initial_prompt and selected_folder.default_initial_prompt:
- initial_prompt = selected_folder.default_initial_prompt
- if not transcription_model and selected_folder.default_transcription_model:
- transcription_model = selected_folder.default_transcription_model
-
- # Apply tag defaults if tags are selected and values are not explicitly provided by user
- # Use first tag's defaults (highest priority - overrides folder)
- if selected_tags:
- first_tag = selected_tags[0]
- if not language and first_tag.default_language:
- language = first_tag.default_language
- if min_speakers is None and first_tag.default_min_speakers:
- min_speakers = first_tag.default_min_speakers
- if max_speakers is None and first_tag.default_max_speakers:
- max_speakers = first_tag.default_max_speakers
- if not hotwords and first_tag.default_hotwords:
- hotwords = first_tag.default_hotwords
- if not initial_prompt and first_tag.default_initial_prompt:
- initial_prompt = first_tag.default_initial_prompt
- if not transcription_model and first_tag.default_transcription_model:
- transcription_model = first_tag.default_transcription_model
-
- # Apply environment variable defaults if still no values are set
- if min_speakers is None and ASR_MIN_SPEAKERS:
- try:
- min_speakers = int(ASR_MIN_SPEAKERS)
- except (ValueError, TypeError):
- min_speakers = None
- if max_speakers is None and ASR_MAX_SPEAKERS:
- try:
- max_speakers = int(ASR_MAX_SPEAKERS)
- except (ValueError, TypeError):
- max_speakers = None
-
- # Fall back to user defaults if still not set
- if not language and current_user.transcription_language:
- language = current_user.transcription_language
- current_app.logger.info(f"Using user's default transcription language: {language}")
- if not hotwords and current_user.transcription_hotwords:
- hotwords = current_user.transcription_hotwords
- if not initial_prompt and current_user.transcription_initial_prompt:
- initial_prompt = current_user.transcription_initial_prompt
+ # Resolve the full transcribe param set through the shared chain used by
+ # every ingestion path: per-request form value > tag > folder > env >
+ # account default (see resolve_transcription_params). Folder defaults
+ # fill any gaps a tag left open (rather than being ignored when a tag is
+ # present), and an empty language falls through the chain rather than
+ # forcing auto-detect. The resolver also applies the admin-curated model
+ # validation/default, so no separate _resolve_transcription_model call is
+ # needed below.
+ resolved_params = resolve_transcription_params(
+ overrides={
+ 'language': language,
+ 'min_speakers': min_speakers,
+ 'max_speakers': max_speakers,
+ 'hotwords': hotwords,
+ 'initial_prompt': initial_prompt,
+ 'transcription_model': transcription_model,
+ },
+ tags=selected_tags,
+ folder=selected_folder,
+ owner=current_user,
+ )
# Create initial database entry
now = datetime.utcnow()
@@ -2848,21 +2814,9 @@ def upload_file():
current_app.logger.info(f"Initial recording record created with ID: {recording.id}")
- # Validate against admin-curated list and apply admin default when
- # nothing else in the chain set a model.
- transcription_model = _resolve_transcription_model(transcription_model)
-
# --- Queue transcription job ---
- first_tag = selected_tags[0] if selected_tags else None
- job_params = {
- 'language': language,
- 'min_speakers': min_speakers,
- 'max_speakers': max_speakers,
- 'tag_id': first_tag.id if first_tag else None,
- 'hotwords': hotwords,
- 'initial_prompt': initial_prompt,
- 'transcription_model': transcription_model,
- }
+ # Params were already resolved (incl. admin model validation) above.
+ job_params = resolved_params
current_app.logger.info(f"Queueing transcription for recording {recording.id} with params: {job_params}")
job_queue.enqueue(
diff --git a/src/services/transcription_defaults.py b/src/services/transcription_defaults.py
index b7f56ecb..4441d318 100644
--- a/src/services/transcription_defaults.py
+++ b/src/services/transcription_defaults.py
@@ -86,22 +86,26 @@ def resolve_transcription_model(value):
return default or None
-def resolve_transcription_params(recording, overrides=None):
+def resolve_transcription_params(recording=None, overrides=None, *, tags=None, folder=None, owner=None):
"""Resolve the complete transcribe job params for a recording.
Args:
- recording: the Recording (provides .tags / .tag_associations / .folder /
- .owner for the default chain).
+ recording: the Recording, when one already exists — its .tags / .folder /
+ .owner supply the default chain. May be omitted by callers that
+ resolve BEFORE the row exists (upload), which instead pass ``tags``,
+ ``folder`` and ``owner`` explicitly.
overrides: optional dict of already-parsed per-request values. Recognized
keys: ``language``, ``min_speakers``, ``max_speakers``, ``hotwords``,
- ``initial_prompt``, ``transcription_model``, ``tag_id``. Any key that
- is absent or None/empty is filled from the recording's tag -> folder
- -> env -> owner defaults.
-
- ``language`` is special: if the key is PRESENT (even as '' meaning
- auto-detect) it is used verbatim; if ABSENT, the owner's default
- language is used. ``tag_id`` is special: if the key is PRESENT (even
- as None) it is respected; if ABSENT, the first tag's id is used.
+ ``initial_prompt``, ``transcription_model``, ``tag_id``. Any value
+ that is absent or None/empty falls through the tag -> folder -> env
+ -> owner chain. ``language`` follows the SAME rule as every other
+ field: a real value wins, an empty/blank value is not "auto-detect"
+ but simply unset and falls through the chain (auto-detect only
+ results when nothing in the chain sets a language). ``tag_id`` is
+ special: if the key is PRESENT (even as None) it is respected; if
+ ABSENT, the first tag's id is used.
+ tags / folder / owner: explicit context, used when ``recording`` is not
+ passed. Ordered ``tags`` list; ``folder`` and ``owner`` objects.
Returns:
dict with keys language, min_speakers, max_speakers, tag_id, hotwords,
@@ -112,6 +116,17 @@ def resolve_transcription_params(recording, overrides=None):
overrides = overrides or {}
+ # Derive context from the recording when given; otherwise the caller (upload,
+ # which resolves before the row exists) supplies tags/folder/owner directly.
+ if recording is not None:
+ if tags is None:
+ tags = recording.tags
+ if folder is None:
+ folder = recording.folder
+ if owner is None:
+ owner = recording.owner
+ tags = list(tags) if tags else []
+
def _override_str(key):
v = overrides.get(key)
if isinstance(v, str):
@@ -123,39 +138,29 @@ def resolve_transcription_params(recording, overrides=None):
hotwords = _override_str('hotwords')
initial_prompt = _override_str('initial_prompt')
transcription_model = _override_str('transcription_model')
-
- owner = recording.owner
-
- # language: an explicit override key (including '' for auto-detect) wins and
- # skips the default chain entirely. Otherwise language resolves through the
- # same tag -> folder -> owner chain as every other field.
- language_explicit = 'language' in overrides
- language = overrides.get('language') if language_explicit else None
-
- # Tag defaults — first tag (by association order) that supplies each value.
- first_tag = None
- if recording.tags:
- for assoc in sorted(recording.tag_associations, key=lambda x: x.order):
- tag = assoc.tag
- if first_tag is None:
- first_tag = tag
- if not language_explicit and not language and tag.default_language:
- language = tag.default_language
- if min_speakers is None and tag.default_min_speakers:
- min_speakers = tag.default_min_speakers
- if max_speakers is None and tag.default_max_speakers:
- max_speakers = tag.default_max_speakers
- if not hotwords and tag.default_hotwords:
- hotwords = tag.default_hotwords
- if not initial_prompt and tag.default_initial_prompt:
- initial_prompt = tag.default_initial_prompt
- if not transcription_model and tag.default_transcription_model:
- transcription_model = tag.default_transcription_model
+ # language behaves like every other field: a real override wins; an empty or
+ # blank value falls through the tag -> folder -> owner chain.
+ language = _override_str('language')
+
+ # Tag defaults — first tag (by order) that supplies each value.
+ first_tag = tags[0] if tags else None
+ for tag in tags:
+ if not language and tag.default_language:
+ language = tag.default_language
+ if min_speakers is None and tag.default_min_speakers:
+ min_speakers = tag.default_min_speakers
+ if max_speakers is None and tag.default_max_speakers:
+ max_speakers = tag.default_max_speakers
+ if not hotwords and tag.default_hotwords:
+ hotwords = tag.default_hotwords
+ if not initial_prompt and tag.default_initial_prompt:
+ initial_prompt = tag.default_initial_prompt
+ if not transcription_model and tag.default_transcription_model:
+ transcription_model = tag.default_transcription_model
# Folder defaults.
- folder = recording.folder
if folder:
- if not language_explicit and not language and folder.default_language:
+ if not language and folder.default_language:
language = folder.default_language
if min_speakers is None and folder.default_min_speakers:
min_speakers = folder.default_min_speakers
@@ -176,7 +181,7 @@ def resolve_transcription_params(recording, overrides=None):
# Owner (account-level) defaults.
if owner:
- if not language_explicit and not language and owner.transcription_language:
+ if not language and owner.transcription_language:
language = owner.transcription_language
if not hotwords and owner.transcription_hotwords:
hotwords = owner.transcription_hotwords
diff --git a/tests/test_transcription_defaults.py b/tests/test_transcription_defaults.py
index 55466c2d..ab69942e 100644
--- a/tests/test_transcription_defaults.py
+++ b/tests/test_transcription_defaults.py
@@ -94,21 +94,37 @@ def test_explicit_override_beats_everything():
assert p["min_speakers"] == 4 # coerced from string
-def test_explicit_empty_language_is_autodetect_not_owner_default():
+def test_empty_language_falls_through_to_defaults():
with app.app_context():
user = _mk_user(transcription_language="fr")
rec = _mk_recording(user)
- # 'language' present but empty => auto-detect, must NOT fall back to 'fr'
- p = resolve_transcription_params(rec, {"language": ""})
- assert p["language"] == ""
+ # An empty/blank language is NOT a terminal auto-detect; it falls through
+ # the chain to the account default 'fr'.
+ assert resolve_transcription_params(rec, {"language": ""})["language"] == "fr"
+ assert resolve_transcription_params(rec, {"language": " "})["language"] == "fr"
+ # Absent language behaves the same.
+ assert resolve_transcription_params(rec, {"hotwords": "x"})["language"] == "fr"
-def test_absent_language_falls_back_to_owner_default():
+def test_auto_detect_only_when_no_default_anywhere():
with app.app_context():
- user = _mk_user(transcription_language="fr")
+ user = _mk_user() # no account default language
rec = _mk_recording(user)
- p = resolve_transcription_params(rec, {"hotwords": "x"}) # no language key
- assert p["language"] == "fr"
+ # Nothing in the chain sets a language => auto-detect (None).
+ assert resolve_transcription_params(rec, {"language": ""})["language"] is None
+
+
+def test_resolver_accepts_explicit_tags_folder_owner_without_recording():
+ # Upload resolves before the recording row exists.
+ with app.app_context():
+ user = _mk_user(transcription_hotwords="acct_hw")
+ tag = Tag(name=f"t_{uuid.uuid4().hex[:6]}", user_id=user.id, default_language="de")
+ db.session.add(tag)
+ db.session.commit()
+ p = resolve_transcription_params(tags=[tag], folder=None, owner=user)
+ assert p["language"] == "de"
+ assert p["hotwords"] == "acct_hw"
+ assert p["tag_id"] == tag.id
def test_explicit_tag_id_none_is_respected():