privacy. Get consent before recording and transcribing other people.
Who it is forSomeone writing up meeting or interview notes.
The problemPlain transcripts do not say who spoke when.
The outcomeJSON or printed segments with start times, a speaker index and text.
What you need first
- Python >= 3.8 with PyTorch and torchaudio installed first, then `pip install funasr`
- A CPU is enough for SenseVoiceSmall; use device cuda only after confirming torch.cuda.is_available() prints True
- Python 3.11 plus fastapi, uvicorn and python-multipart for the local HTTP server
Steps
- Install with
pip install torch torchaudioandpip install funasr - For a quick result run
funasr audio.wav --spk --timestamps -f json - For more control, use the README's Python example: AutoModel with model iic/SenseVoiceSmall, vad_model fsmn-vad and spk_model cam++, then loop over sentence_info
- Rename the anonymous speaker indices to real names yourself after reviewing
Prompt for your agent
With modelscope/FunASR, run `funasr audio.wav --spk --timestamps -f json` on <my meeting recording> and turn the JSON into a readable transcript grouped by speaker.
Caveat. The README says speaker indices are local to each recording, not known person identities.
Use this in your AI agent
Ask your agent
Using the Deduped MCP, get the setup steps for modelscope/FunASR and "Transcribe a meeting with speaker labels"
Add Deduped to Claude Code once
claude mcp add --transport http deduped https://dedupedai.com/mcp
Every Monday: the 5 hottest repos and exactly how to use them. Free, one email a week, unsubscribe anytime.
Get them freeShare: X LinkedIn Hacker News Email Copy link
Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer