BUT-FIT/diarizen-meeting-base

Overview

This hub features the pre-trained model by DiariZen. The EEND component is built upon WavLM-Base+ and Conformer layers. The model was trained on far-field, single-channel audio data from the public datasets AMI, AISHELL-4, and AliMeeting. Please follow the instructions for before use.

Usage

from diarizen.pipelines.inference import DiariZenPipeline

# load pre-trained model
diar_pipeline = DiariZenPipeline.from_pretrained("BUT-FIT/diarizen-meeting-base")
# apply diarization pipeline
diar_results = diar_pipeline('audio.wav')

# print results
for turn, _, speaker in diar_results.itertracks(yield_label=True):
    print(f"start={turn.start:.1f}s stop={turn.end:.1f}s speaker_{speaker}")

# load pre-trained model and save RTTM result
diar_pipeline = DiariZenPipeline.from_pretrained(
        "BUT-FIT/diarizen-meeting-base",
        rttm_out_dir='.'
)
# apply diarization pipeline
diar_results = diar_pipeline('audio.wav', sess_name='session_name')

Results (SDM)

Diarization Error Rate (DER):
--------------------------------------------------------------
System         Collar       AMI   AISHELL-4   AliMeeting         
--------------------------------------------------------------
Pyannote3        0s         21.1     13.9       22.8
                0.25s       13.7      7.7       13.6
--------------------------------------------------------------
Proposed         0s         15.4     11.7       17.6
                0.25s        9.8      5.9       10.2
--------------------------------------------------------------

Citation

@inproceedings{han2025leveraging,
  title={Leveraging self-supervised learning for speaker diarization},
  author={Han, Jiangyu and Landini, Federico and Rohdin, Johan and Silnova, Anna and Diez, Mireia and Burget, Luk{\'a}{\v{s}}},
  booktitle={Proc. ICASSP},
  year={2025}
}