Skip to content

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

Python

Incognito uses the upload settings chain (#412 audit P14): user language for an empty value, account and admin hotwords and initial prompt, ASR speaker-count defaults; title uses the admin transcript limit; the admin auto-summarization switch applies

Murtaza Nasirdb9dfc6ded31
diff --git a/src/api/recordings.py b/src/api/recordings.py
index 897497b4..f6fe4234 100644
--- a/src/api/recordings.py
+++ b/src/api/recordings.py
@@ -2944,7 +2944,8 @@ def upload_incognito():
current_app.logger.info(f"[Incognito] Temp file saved: {temp_filepath}")
# Get optional parameters
- # Note: Empty string '' means auto-detect, don't convert to None
+ # An empty language falls through to the user's default, as on upload;
+ # "auto" forces auto-detection (resolved in transcribe_incognito, #412).
language = request.form.get('language', '')
min_speakers = request.form.get('min_speakers')
max_speakers = request.form.get('max_speakers')
@@ -2952,6 +2953,10 @@ def upload_incognito():
initial_prompt = request.form.get('initial_prompt', '').strip() or None
transcription_model = request.form.get('transcription_model', '').strip() or None
auto_summarize = request.form.get('auto_summarize', 'false').lower() == 'true'
+ # The admin switch that disables automatic summaries applies here too (#412).
+ _admin_off = SystemSetting.get_setting('disable_auto_summarization', False)
+ if (_admin_off if isinstance(_admin_off, bool) else str(_admin_off).lower() == 'true'):
+ auto_summarize = False
# Convert to int if provided
if min_speakers:
diff --git a/src/tasks/processing.py b/src/tasks/processing.py
index 9b7a8c6a..703888f6 100644
--- a/src/tasks/processing.py
+++ b/src/tasks/processing.py
@@ -2617,9 +2617,21 @@ def transcribe_incognito(filepath, original_filename, language=None, min_speaker
# Determine diarization settings (respects ASR_DIARIZE env var)
should_diarize = getattr(connector, 'default_diarize', connector.supports_diarization)
- # Use user's language preference if not explicitly provided
- if language is None and user:
- language = user.transcription_language
+ # Same settings chain as an upload (#412 audit P14): request value, then
+ # ASR speaker-count env defaults, the user's defaults and the admin
+ # defaults. Incognito has no tags or folder. An empty language falls
+ # through to the user's language, as on upload ("auto" forces detection).
+ from src.services.transcription_defaults import resolve_transcription_params
+ _resolved = resolve_transcription_params(
+ None,
+ {'language': language, 'min_speakers': min_speakers, 'max_speakers': max_speakers,
+ 'hotwords': hotwords, 'initial_prompt': initial_prompt,
+ 'transcription_model': transcription_model},
+ tags=[], folder=None, owner=user)
+ language = _resolved['language']
+ min_speakers, max_speakers = _resolved['min_speakers'], _resolved['max_speakers']
+ hotwords, initial_prompt = _resolved['hotwords'], _resolved['initial_prompt']
+ transcription_model = _resolved['transcription_model']
# Normalize at the boundary — legacy values like "français" must
# become "fr" before the connector receives them (issue #256).
@@ -2636,9 +2648,6 @@ def transcribe_incognito(filepath, original_filename, language=None, min_speaker
# Same model an upload would get, so a service without its own default
# model works in incognito mode too (#409).
- from src.services.transcription_defaults import resolve_transcription_model
- transcription_model = resolve_transcription_model(transcription_model)
-
current_app.logger.info(f"[Incognito] Starting transcription: diarize={should_diarize}, language={language}, chunking={should_chunk}")
if should_chunk:
@@ -2718,7 +2727,9 @@ def _generate_incognito_title(transcription_text, user=None):
# Get formatted text for LLM
formatted_text = format_transcription_for_llm(transcription_text)
# Limit text for title generation
- limited_text = formatted_text[:5000]
+ # Same transcript limit as the stored title (admin setting, #412).
+ _limit = SystemSetting.get_setting('transcript_length_limit', 30000)
+ limited_text = formatted_text if _limit == -1 else formatted_text[:_limit]
# Get user language preference
user_output_language = user.output_language if user else None
diff --git a/tests/test_incognito_settings.py b/tests/test_incognito_settings.py
new file mode 100644
index 00000000..bcb4608d
--- /dev/null
+++ b/tests/test_incognito_settings.py
@@ -0,0 +1,69 @@
+"""Incognito uses the same settings chain as an upload (#412 audit P14).
+
+It used to pass the form's empty language straight through (so the user's
+language never applied), and it skipped the account and admin hotwords and
+initial prompt and the ASR speaker-count defaults. Nothing is stored either way.
+
+SHARED-DB: the user and admin setting are restored afterwards.
+"""
+
+import os
+import sys
+import uuid
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+from src.app import app, db
+from src.models import SystemSetting, User
+import src.tasks.processing as proc
+
+
+def user():
+ with app.app_context():
+ old = SystemSetting.get_setting("admin_default_initial_prompt", None)
+ SystemSetting.set_setting("admin_default_initial_prompt", "A planning meeting.", setting_type="string")
+ s = uuid.uuid4().hex[:8]
+ u = User(username=f"inc_{s}", email=f"inc_{s}@local.test", password="x",
+ transcription_language="de", transcription_hotwords="Speakr")
+ db.session.add(u)
+ db.session.commit()
+ uid = u.id
+ yield uid
+ with app.app_context():
+ if old is None:
+ SystemSetting.query.filter_by(key="admin_default_initial_prompt").delete()
+ else:
+ SystemSetting.set_setting("admin_default_initial_prompt", old, setting_type="string")
+ db.session.delete(db.session.get(User, uid))
+ db.session.commit()
+
+
+def test_incognito_resolves_language_hotwords_and_admin_prompt(user, tmp_path):
+ audio = tmp_path / "a.mp3"
+ audio.write_bytes(b"\x00" * 2048)
+ connector = MagicMock(PROVIDER_NAME="test", supports_diarization=False, default_diarize=False)
+ connector.specifications = MagicMock(max_duration_seconds=1800, handles_chunking_internally=False,
+ recommended_chunk_seconds=600, unsupported_codecs=[])
+ response = MagicMock(text="hello world transcript", segments=None, speakers=None, speaker_embeddings=None)
+ response.has_diarization.return_value = False
+ connector.transcribe.return_value = response
+ with app.app_context():
+ u = db.session.get(User, user)
+ registry = MagicMock()
+ registry.get_active_connector.return_value = connector
+ with patch("src.services.transcription.get_registry", return_value=registry), \
+ patch.object(proc, "is_video_file", return_value=False), \
+ patch.object(proc, "convert_if_needed", return_value=MagicMock(was_converted=False)), \
+ patch.object(proc, "chunking_service") as svc, patch.object(proc, "client", None):
+ svc.needs_chunking.return_value = False
+ svc.get_audio_duration.return_value = 30.0
+ proc.transcribe_incognito(str(audio), "a.mp3", language="", user=u)
+ call = connector.transcribe.call_args
+ request = call.args[0] if call.args else next(iter(call.kwargs.values()))
+ assert request.language == "de"
+ assert request.hotwords == "Speakr"
+ assert request.prompt == "A planning meeting."