ComfyUI Speaker Diarization Node (pyannote)

Custom ComfyUI node using pyannote-audio speaker diarization.

Inputs

• audio (AUDIO): ComfyUI audio input.
• hf_token (STRING): Hugging Face access token.
• whisper_model: Whisper model size (tiny, base, small, medium, large, turbo).
• merge_consecutive_speaker (BOOLEAN): merge consecutive lines when the same speaker keeps talking.

Output

• diarization_text (STRING) in format:
  • 0:10 SPEAKER A: ...
  • 0:20 SPEAKER B: ...

Each line is a Whisper transcription segment tagged with the speaker that overlaps most with the segment.

Install

1. Install dependencies:

   pip install -r requirements.txt
   

2. Restart ComfyUI.

Notes

• You must have access to pyannote/speaker-diarization-3.1 on Hugging Face with your token.
• First run will download both pyannote and Whisper models.