Feature Request: Add FunASR/SenseVoice as a self-hosted STT provider

[LauraGPT]

Dograh's modular STT architecture is a great fit for adding FunASR as a self-hosted STT provider. This would give users a completely self-contained voice AI stack with no external API dependencies.

Why FunASR?

• OpenAI-compatible API: FunASR ships with funasr-server that serves at /v1/audio/transcriptions — same API as OpenAI Whisper, so integration is minimal
• 170x realtime on GPU, 17x on CPU — ideal for real-time voice agent pipelines
• 50+ languages with strong CJK support
• Built-in VAD + punctuation — no need for separate voice activity detection
• Speaker diarization — cam++ model included for multi-speaker scenarios
• Streaming support — Paraformer-streaming for low-latency real-time ASR

Integration

# Start FunASR server (OpenAI-compatible)
pip install funasr
funasr-server --device cuda
# Serves at localhost:8000/v1/audio/transcriptions

Since Dograh already supports custom STT providers, pointing to the FunASR server endpoint should work with minimal code changes.

Resources

• FunASR: https://github.com/modelscope/FunASR (16.6K+ stars)
• SenseVoice: https://github.com/FunAudioLLM/SenseVoice (8.3K+ stars)

This would make Dograh one of the few voice AI platforms with a truly zero-external-dependency option for STT.