setup_diarization.md

Setting Up Speaker Diarization

Speaker diarization requires a HuggingFace token to download the pyannote models.

Steps:

1. Get a HuggingFace Token:

   • Go to: https://huggingface.co/settings/tokens
   • Create a new token (read access is sufficient)
   • Copy the token

2. Accept the pyannote model license:

   • Go to: https://huggingface.co/pyannote/speaker-diarization-3.1
   • Click "Agree and access repository"
   • Also visit: https://huggingface.co/pyannote/segmentation-3.0
   • Click "Agree and access repository"

3. Set the token:

   export HF_TOKEN="your_token_here"

4. Re-run the transcription:

   source .venv/bin/activate
   python examples/transcribe_only.py ~/Movies/"2025-11-04 14-36-25.mp4" \
     --model tiny \
     --output-dir output

The transcription will now include speaker labels like SPEAKER_00, SPEAKER_01, etc.