For more information on how to run this diarization model see https://github.com/revdotcom/reverb/tree/main/diarization

Reverb diarization V2 provides a 22.25% relative improvement in WDER (Word Diarization Error Rate) compared to the baseline pyannote3.0 model, evaluated on over 1,250,000 tokens across five different test suites.

Test suite	WDER
earnings21	0.046
rev16	0.078

Usage

# taken from https://huggingface.co/pyannote/speaker-diarization-3.1 - see for more details
# instantiate the pipeline
from pyannote.audio import Pipeline
pipeline = Pipeline.from_pretrained(
  "Revai/reverb-diarization-v2",
  use_auth_token="HUGGINGFACE_ACCESS_TOKEN_GOES_HERE")

# run the pipeline on an audio file
diarization = pipeline("audio.wav")

# dump the diarization output to disk using RTTM format
with open("audio.rttm", "w") as rttm:
    diarization.write_rttm(rttm)

License

See LICENSE for details.