Skip to content

feat(ffmpeg-whisper): add provider selection and gpu auto-detect #44

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
POWERFULMOVES merged 1 commit into from
Sep 20, 2025
Merged

feat(ffmpeg-whisper): add provider selection and gpu auto-detect #44

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 3 additions & 2 deletions pmoves/docs/PMOVES.yt/PMOVES_YT.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,7 +3,7 @@
Endpoints
- POST `/yt/info`: { url } → returns { id, title, uploader, duration, webpage_url }.
- POST `/yt/download`: { url, bucket?, namespace? } → downloads MP4, uploads to S3 at `yt/<id>/raw.mp4`, optional thumbnail; inserts `studio_board` + `videos`; emits `ingest.file.added.v1`.
- POST `/yt/transcript`: { video_id, bucket?, namespace?, language?, whisper_model? } → extracts audio + runs Whisper (via ffmpeg‑whisper); inserts `transcripts`; emits `ingest.transcript.ready.v1`.
- POST `/yt/transcript`: { video_id, bucket?, namespace?, language?, whisper_model?, provider? } → extracts audio + runs the selected transcription backend (`faster-whisper` default, `whisper`, or `qwen2-audio`); inserts `transcripts`; emits `ingest.transcript.ready.v1`.
- POST `/yt/ingest`: convenience: info + download + transcript.
- POST `/yt/playlist`: { url, namespace?, bucket?, max_videos?, … } → iterates playlist, tracks job state in `yt_jobs/yt_items`, downloads + transcribes each video.
- POST `/yt/channel`: { url|channel_id, ... } → same as playlist for a channel.
Expand Down Expand Up @@ -44,7 +44,8 @@ Flow
5) Downstream: LangExtract → extract-worker → Hi‑RAG + Neo4j; Jellyfin refresh + Discord embeds (pending)

Notes
- ffmpeg‑whisper currently uses OpenAI Whisper; GPU path is to switch to `faster-whisper`.
- `ffmpeg-whisper` auto-detects CUDA GPUs (`WHISPER_DEVICE` / `USE_CUDA` override) and defaults to `faster-whisper`, falling back to CPU INT8 inference when no GPU is available.
- Set `provider` in `/yt/transcript` or `/yt/ingest` to choose `faster-whisper`, `whisper` (WhisperX + optional PyAnnote diarization), or `qwen2-audio` (requires Transformers + Qwen2 Audio weights; GPU strongly recommended for real-time throughput).
- For Jetson, use L4T PyTorch bases and CUDA‑enabled models.
- Gemma providers:
- `YT_SUMMARY_PROVIDER=ollama|hf`, `OLLAMA_URL`, `YT_GEMMA_MODEL` (default gemma2:9b-instruct)
Expand Down
3 changes: 3 additions & 0 deletions pmoves/services/ffmpeg-whisper/requirements.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,3 +8,6 @@ onnxruntime==1.19.2
boto3==1.34.162
requests==2.32.3
supabase==2.3.1
faster-whisper==1.0.2
transformers==4.44.2
sentencepiece==0.2.0
Loading
Loading