diarizen-wavlm-large-s80-mlc

Overview

This hub features the pre-trained model by DiariZen as described in BUT System for the MLC-SLM Challenge. The EEND component is built upon WavLM-Large and Conformer layers. The model was pre-trained on far-field, single-channel audio from a diverse set of public datasets, including AMI, AISHELL-4, AliMeeting, NOTSOFAR-1, MSDWild, DIHARD3, RAMC, and VoxConverse. Then structured pruning at 80% sparsity is applied. Finally, the pruned model is fine-tuned with MLC-SLM data. When loading this model, please ensure non-commercial usage, in accordance with the CC BY-NC 4.0 license.

Usage

from diarizen.pipelines.inference import DiariZenPipeline

# load pre-trained model
diar_pipeline = DiariZenPipeline.from_pretrained("BUT-FIT/diarizen-wavlm-large-s80-mlc")
# 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-wavlm-large-s80-mlc",
        rttm_out_dir='.'
)
# apply diarization pipeline
diar_results = diar_pipeline('audio.wav', sess_name='session_name')

Results

DER evaluation of Pyannote baseline and DiariZen, with no collar applied.

Citation

If you found this work helpful, please consider citing:

@article{polok2025but,
  title={BUT System for the MLC-SLM Challenge},
  author={Polok, Alexander and Han, Jiangyu and Klement, Dominik and Cornell, Samuele and {\v{C}}ernock{\`y}, Jan and Burget, Luk{\'a}{\v{s}}},
  journal={arXiv preprint arXiv:2506.13414},
  year={2025}
}

License

• Source code: MIT (see the project’s GitHub repository).
• Model weights: CC BY-NC 4.0 (non-commercial).
• Rationale: some training datasets are research-only or non-commercial, so the released weights cannot be used commercially.