yap-diarize-models — Yap's pinned diarization model mirror

This repository is a byte-for-byte mirror of two upstream sherpa-onnx release assets. Nothing here is re-exported, re-quantized or re-packaged: the files are uploaded exactly as the upstream release serves them, so their sha256 digests match the upstream digests.

It exists so that Yap's models.rs catalog can pin a revision + sha256 it controls, instead of downloading vendor release archives directly at runtime.

File	Bytes	sha256	Upstream
sherpa-onnx-pyannote-segmentation-3-0.tar.bz2	6,958,444	24615ee884c897d9d2ba09bb4d30da6bb1b15e685065962db5b02e76e4996488	k2-fsa/sherpa-onnx speaker-segmentation-models
wespeaker_en_voxceleb_CAM++.onnx	29,292,684	c46fad10b5f81e1aa4a60c162714208577093655076c5450f8c469e522ec54ef	k2-fsa/sherpa-onnx speaker-recongition-models (upstream tag typo is upstream's)

Licenses (unchanged by mirroring)

• pyannote/segmentation-3.0 — MIT (LICENSE inside the archive, © 2022 CNRS). The model card states it will remain open-source. The archive also carries an model.int8.onnx quantized sibling and the upstream export scripts; Yap loads sherpa-onnx-pyannote-segmentation-3-0/model.onnx (5,992,913 bytes, sha256 220ad67ca923bef2fa91f2390c786097bf305bceb5e261d4af67b38e938e1079).
• wespeaker_en_voxceleb_CAM++ — WeSpeaker toolkit is Apache-2.0; the VoxCeleb-derived weights are CC-BY-4.0. 192-dimensional embeddings.

Rev's sherpa-onnx-reverb-diarization-v{1,2} models are deliberately not mirrored here: they are released for non-commercial use only, which is incompatible with a paid product.