Feature: Add SenseVoice with an optional diarization pipeline

[LauraGPT]

Important

Correction: The earlier comparative claims below are withdrawn; they were not established by a matched benchmark. SenseVoiceSmall supports Mandarin Chinese, Cantonese, English, Japanese, and Korean. It can emit language, emotion, and audio-event tags, but speaker diarization requires a separate model or pipeline (for example CAM++) and is not a built-in SenseVoice result. Runtime, timestamps, punctuation, and performance depend on the selected model, interface, hardware, and audio. FunASR and SenseVoice repository source code is MIT; model weights follow each model card. Please evaluate the exact integration on this project's workload.

Hi! Great project — fully local meeting transcription with speaker diarization and AI summaries.

SenseVoice + cam++ from the FunASR toolkit could be a strong addition:

Why it fits Millet

• 5x faster than Whisper — non-autoregressive architecture
• Built-in speaker diarization — cam++ model (7.2M params, SOTA accuracy)
• 234M params — runs on CPU, ideal for local deployment
• Emotion detection — useful for meeting sentiment analysis
• PDF support via word-level timestamps — clean transcript formatting

One-line pipeline

from funasr import AutoModel

model = AutoModel(
    model="iic/SenseVoiceSmall",
    vad_model="fsmn-vad",
    spk_model="cam++",
)
result = model.generate(input="meeting.wav")

Links

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