Optional speaker diarization — pinned distribution and notices, reviewed 2026-10-21.

Binary/runtime/model payload (UNCHANGED):
https://github.com/k2-fsa/sherpa-onnx/releases/download/v1.13.7/sherpa-onnx-wasm-simd-v1.13.7-speaker-diarization.tar.bz2
SHA256 f197725498893010a8da8f18c62ee67c62ed7294ab33d68a0b182211fcba0b7d
4-file catalog: src/models/diarization-catalog.json.
segmentation.onnx SHA256 220ad67ca923bef2fa91f2390c786097bf305bceb5e261d4af67b38e938e1079
embedding.onnx SHA256 1a331345f04805badbb495c775a6ddffcdd1a732567d5ec8b3d5749e3c7a5e4b
Segmentation permission is the original MIT LICENSE in the model's official release archive (CNRS, 2022). 3D-Speaker's model metadata designates Apache License 2.0:
https://modelscope.cn/models/iic/speech_eres2net_base_sv_zh-cn_3dspeaker_16k
https://modelscope.cn/api/v1/models/iic/speech_eres2net_base_sv_zh-cn_3dspeaker_16k

Source code, build instructions and third-party originals (NO CHANGES by this app):
sherpa-onnx 1.13.7, Copyright Xiaomi Corporation and contributors, Apache-2.0:
https://github.com/k2-fsa/sherpa-onnx/archive/refs/tags/v1.13.7.tar.gz
https://github.com/k2-fsa/sherpa-onnx/blob/v1.13.7/build-wasm-simd-speaker-diarization.sh
https://github.com/k2-fsa/sherpa-onnx/blob/v1.13.7/.github/workflows/wasm-simd-hf-space-speaker-diarization.yaml
ONNX Runtime 1.27.1, MIT, original aggregate ThirdPartyNotices included:
https://github.com/microsoft/onnxruntime/archive/refs/tags/v1.27.1.tar.gz
https://github.com/csukuangfj/onnxruntime-libs/releases/download/v1.27.1/onnxruntime-wasm-static_lib-simd-1.27.1.zip
kaldi-native-fbank 1.22.3, Apache-2.0: https://github.com/csukuangfj/kaldi-native-fbank/archive/refs/tags/v1.22.3.tar.gz
KISS FFT febd4caeed32e33ad8b2e0bb5ea77542c40f18ec, BSD-3-Clause:
https://github.com/mborgerding/kissfft/archive/febd4caeed32e33ad8b2e0bb5ea77542c40f18ec.zip
hclust-cpp 2026-02-25, BSD-2-Clause: https://github.com/csukuangfj/hclust-cpp/archive/refs/tags/2026-02-25.tar.gz
Eigen 5.0.1, primarily MPL-2.0. Original complete source is available at either location, including all covered files and file-level notices:
https://gitlab.com/libeigen/eigen/-/archive/5.0.1/eigen-5.0.1.tar.gz
https://github.com/eigen-mirror/eigen/archive/refs/tags/5.0.1.tar.gz
The accompanying Eigen COPYING.* files cover its unmodified source distribution. No GPL/LGPL optional Eigen routines are selected by this speaker entry point.
Linked build dependencies retained conservatively (not a claim that every object survives linker GC):
kaldi-decoder 0.3.0, Apache-2.0: https://github.com/k2-fsa/kaldi-decoder/archive/refs/tags/v0.3.0.tar.gz
kaldifst 1.8.0, Apache-2.0: https://github.com/k2-fsa/kaldifst/archive/refs/tags/v1.8.0.tar.gz
OpenFst 1.8.5-2026-04-10, Apache-2.0: https://github.com/csukuangfj/openfst/archive/refs/tags/v1.8.5-2026-04-10.tar.gz
simple-sentencepiece 0.7, Apache-2.0: https://github.com/pkufool/simple-sentencepiece/archive/refs/tags/v0.7.tar.gz
nlohmann/json 3.12.0, MIT: https://github.com/nlohmann/json/archive/refs/tags/v3.12.0.tar.gz
Emscripten 4.0.23, MIT/UIUC, with libc++/libc++abi Apache-2.0 WITH LLVM-exception and musl MIT/per-file notices:
https://github.com/emscripten-core/emscripten/archive/refs/tags/4.0.23.tar.gz

The executable exports only the offline speaker-diarization C API and allocator functions. TTS/phonemizer models, dictionaries, browser demos and unrelated upstream entry points are not distributed or executed. Upstream file hashes are checked before copying and before browser execution. The authored bridge does not modify upstream JS or models.
All fetched notice originals and their SHA256/source URLs are recorded in SOURCE.json. Explanatory text in this file is authored by this application; original license and notice files are unmodified.