Oliver/video-accessibility
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

fix: run Whisper transcription inline instead of as subtask

Browse source 
Celery does not allow calling result.get() within a task as it causes
deadlocks. Changed the implementation to run Whisper transcription
directly using asyncio.to_thread() instead of dispatching to a separate
Celery queue.

The Whisper transcript is still cached in MongoDB for reuse across
language variants.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
michael 2025-12-27 08:48:41 -06:00
parent 3ca70a7c03
commit 7b0ebb357c
1 changed files with 7 additions and 17 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

24
backend/app/tasks/render_accessible_video.py
View file

@ -19,7 +19,6 @@ from ..services.vtt_retimer import vtt_retimer_service
from ..services.whisper_service import WordTimestamp, whisper_service
from . import celery_app
from .translate_and_synthesize import broadcast_status_update, retry_with_backoff
from .whisper_transcribe import transcribe_video_audio_task
logger = get_logger(__name__)
@ -395,32 +394,23 @@ async def _refine_pause_points_with_whisper(
audio_path = os.path.join(temp_dir, "source_audio.mp3")
await _extract_audio_for_whisper(video_path, audio_path)
# Dispatch to Whisper queue and wait for result
logger.info(f"Dispatching Whisper transcription task for job {job_id}")
result = transcribe_video_audio_task.apply_async(
args=[job_id, audio_path],
queue="whisper"
)
# Wait for transcription to complete (up to 30 minutes)
# Run Whisper transcription directly (inline)
# Note: We run this inline instead of via Celery subtask because
# calling result.get() within a task causes deadlocks
logger.info(f"Running Whisper transcription inline for job {job_id}")
try:
transcription_result = result.get(timeout=1800)
# Run Whisper in a thread pool to avoid blocking async loop
words = await asyncio.to_thread(whisper_service.transcribe_audio, audio_path)
except Exception as e:
logger.error(f"Whisper transcription failed for job {job_id}: {e}")
return analysis, [f"Whisper transcription failed: {str(e)} - using original Gemini timestamps"]
# Convert to WordTimestamp objects
words = [
WordTimestamp(word=w["word"], start=w["start"], end=w["end"])
for w in transcription_result.get("words", [])
]
if words:
# Cache the transcript
cache_data = CachedWhisperTranscript(
words=[CachedWordTimestamp(word=w.word, start=w.start, end=w.end) for w in words],
model_name=settings.whisper_model,
audio_duration=transcription_result.get("audio_duration", 0),
audio_duration=0, # We don't track duration when running inline
created_at=datetime.utcnow().isoformat()
)