Feature Request: Add SenseVoice/FunASR as STT provider

[LauraGPT]

Feature Request

Dograh is building a great open-source voice AI platform. Suggesting SenseVoice / FunASR as an additional STT provider option.

Why SenseVoice for voice AI?

• 5x faster than Whisper — non-autoregressive architecture, critical for real-time voice agents
• 50+ languages in a single 234M param model
• Emotion detection — identifies speaker emotions (happy, angry, sad), useful for sentiment-aware agents
• Audio events — detects laughter, applause, music, background noise
• OpenAI-compatible API — funasr-server serves /v1/audio/transcriptions, easy integration
• Streaming support — WebSocket-based real-time streaming with partial results

Self-hosted advantage

FunASR runs entirely locally — perfect for self-hosted voice AI:

pip install funasr vllm
funasr-server --device cuda  # OpenAI-compatible /v1/audio/transcriptions

Or integrate directly:

from funasr import AutoModel
model = AutoModel(model="iic/SenseVoiceSmall", vad_model="fsmn-vad")
result = model.generate(input=audio_bytes)

Resources

• FunASR: https://github.com/modelscope/FunASR (16.6K stars)
• SenseVoice: https://github.com/FunAudioLLM/SenseVoice (8.3K stars)
• OpenAI-compatible server: built-in with funasr-server