Hi! Verbi's modular architecture for experimenting with different STT/LLM/TTS components is excellent.

I'd like to suggest adding SenseVoice as a new STT option. It fits Verbi's modular philosophy well:

Why SenseVoice:

• 5x faster than Whisper large-v3 (234M non-autoregressive model)
• Emotion detection built-in — useful for adjusting assistant behavior
• Audio event detection — laugh, applause, music, cough, etc.
• 50+ languages with strong Chinese/English/Japanese/Korean support
• Simple pip install funasr — no extra dependencies

Integration example:

from funasr import AutoModel
model = AutoModel(model="iic/SenseVoiceSmall")
result = model.generate(input="audio.wav")
text = result[0]["text"]

Or via OpenAI-compatible API:

funasr-server --device cuda
# POST http://localhost:8000/v1/audio/transcriptions

Would be a great addition to the existing Deepgram/AssemblyAI/Groq STT options.

• FunASR: https://github.com/modelscope/FunASR (16K+ stars)
• SenseVoice: https://github.com/FunAudioLLM/SenseVoice (8K+ stars)