DanieClar/stillscript-speaker-diarization-community-1

StillScript mirror of pyannote/speaker-diarization-community-1

This repository is an unmodified redistribution of pyannote/speaker-diarization-community-1, re-hosted so that StillScript Confidential Transcripts can fetch it without each end user having to individually accept the upstream repository's gating conditions.

License, Attribution, and Citation Information

• Title: speaker-diarization-community-1
• Author: Hervé Bredin and the pyannoteAI / pyannote.audio team
• Source: https://huggingface.co/pyannote/speaker-diarization-community-1
• License: Creative Commons Attribution 4.0 International (CC-BY-4.0)

CC-BY-4.0 permits redistribution provided attribution is given and changes are indicated.

Changes made

None to the model itself. The pipeline weights, the config.yaml, and the component sub-models (segmentation/, embedding/, plda/) are byte-for-byte identical to the upstream repository. The only differences are:

• this README (replacing the upstream model card), and
• the removal of the upstream diarization.gif illustration and .gitattributes, neither of which is used at inference time.

No retraining, fine-tuning, quantisation, or conversion of any kind was performed.

Verifying that for yourself

Every shipped file's SHA-256, as redistributed here:

File	SHA-256
config.yaml	5ce2bfa9a938dc132cec1172592d65173cbb8f444ea1e4133f10f9391de155be
segmentation/pytorch_model.bin	7ad24338d844fb95985486eb1a464e32d229f6d7a03c9abe60f978bacf3f816e
embedding/pytorch_model.bin	6f10ff60898a1d185fa22e1d11e0bfa8a92efec811f11bca48cb8cafebefd929
plda/plda.npz	9b77bcd840692710dd3496f62ecfeed8d8e5f002fd991b785079b244eab7d255
plda/xvec_transform.npz	325f1ce8e48f7e55e9c8aa47e05d2766b7c48c4b25b8de8dd751e7a4cc5fbe8f

StillScript pins these same hashes and verifies them after download.

Citations

The original authors ask that the following be cited. Reproduced verbatim from the upstream model card.

1. Speaker segmentation model

@inproceedings{Plaquet23,
  author={Alexis Plaquet and Hervé Bredin},
  title={{Powerset multi-class cross entropy loss for neural speaker diarization}},
  year=2023,
  booktitle={Proc. INTERSPEECH 2023},
}

2. pyannote.audio toolkit

@inproceedings{Bredin23,
  author={Hervé Bredin},
  title={{pyannote.audio 2.1 speaker diarization pipeline: principle, benchmark, and recipe}},
  year=2023,
  booktitle={Proc. INTERSPEECH 2023},
}

Please prefer the upstream repository if you are not using StillScript — it is the canonical source, and using it directly supports the pyannote project.

---

Mirrored by Acutus Consulting for StillScript Confidential Transcripts. Modelling credit: the pyannote.audio / pyannoteAI team.