Hi! Omi is an impressive AI 2nd brain with its extensible STT provider system.
I'd like to suggest adding FunASR (16K+ stars) as a self-hosted STT provider alternative to Deepgram:
Why FunASR for Omi:
- Self-hosted & private: All processing local — no data leaves the device/server
- 170x real-time GPU speed: Handles real-time conversation transcription easily
- Built-in speaker diarization: cam++ model (7.2M params) identifies different speakers — key for Omi's multi-speaker scenarios
- Emotion detection: SenseVoice detects emotions and audio events alongside transcription
- 50+ languages: Automatic language detection via SenseVoice model
- OpenAI-compatible API: Drop-in replacement for cloud APIs
pip install funasr vllm fastapi uvicorn python-multipart
funasr-server --device cuda
# POST http://localhost:8000/v1/audio/transcriptions
This would give privacy-conscious Omi users a fully self-hosted alternative to Deepgram with comparable features (diarization + emotion).
Hi! Omi is an impressive AI 2nd brain with its extensible STT provider system.
I'd like to suggest adding FunASR (16K+ stars) as a self-hosted STT provider alternative to Deepgram:
Why FunASR for Omi:
pip install funasr vllm fastapi uvicorn python-multipart funasr-server --device cuda # POST http://localhost:8000/v1/audio/transcriptionsThis would give privacy-conscious Omi users a fully self-hosted alternative to Deepgram with comparable features (diarization + emotion).