Feature: Add SenseVoice for faster local meeting transcription

[LauraGPT]

Hi! Interesting approach — meeting transcription using just an LLM.

For the speech-to-text step, have you considered SenseVoice? It could complement or replace the LLM-only approach for initial transcription:

SenseVoice advantages

• 5x faster than Whisper — non-autoregressive architecture
• 234M params — lightweight, runs on CPU
• Built-in features: speaker diarization (cam++), emotion detection, audio event classification
• 50+ languages — auto-detects language
• OpenAI-compatible API — drop-in replacement

Quick start

from funasr import AutoModel

model = AutoModel(
    model="iic/SenseVoiceSmall",
    vad_model="fsmn-vad",
    spk_model="cam++",  # speaker diarization
)
result = model.generate(input="meeting.wav")
# Returns: text + speaker labels + timestamps + emotion

The combination of SenseVoice (fast ASR) + your LLM approach (structuring/summarization) could give best of both worlds — accurate transcription with intelligent formatting.

Links

• SenseVoice: https://github.com/FunAudioLLM/SenseVoice (8.3K stars)
• FunASR: https://github.com/modelscope/FunASR (16.7K stars)