polyvoice
Copyright (c) the polyvoice authors
This product is licensed under the MIT License (see LICENSE).
Third-party model artifacts
===========================
polyvoice's source code is MIT. The neural model weights it loads are NOT part
of this repository — they are downloaded at runtime from their upstream sources
under their own licenses (see src/models/manifest.toml). Redistributors who
bundle the weights must comply with those licenses.
VBx PLDA parameters (opt-in `vbx` clusterer)
--------------------------------------------
The PLDA parameters consumed by the VBx clusterer
(plda_{transform,phi_computed,mean1,mean2,lda,mu}.npy) are derived from the
pyannote `speaker-diarization-community-1` pipeline and are matched to the
`pyannote/wespeaker-voxceleb-resnet34-LM` embedder.
Licensed under CC-BY-4.0 (https://creativecommons.org/licenses/by/4.0/).
Attribution: pyannote (Hervé Bredin et al.).
Redistribution of these derived parameters is permitted under CC-BY-4.0 with
attribution to pyannote. The raw inputs are mirrored at the Apache-2.0
`avencera/speakrs` project's `avencera/speakrs-models` model repository.
Third-party Rust crates (opt-in features)
Symphonia (feature `audio-io`)
------------------------------
polyvoice optionally depends on [Symphonia](https://github.com/pdeljanov/Symphonia)
(and its codec/format sub-crates) for multi-format audio decoding (mp3, flac,
ogg/vorbis, m4a/aac, and related containers).
Licensed under the Mozilla Public License 2.0 (MPL-2.0).
Used as an **unmodified** dependency from crates.io only — sources are neither
vendored nor patched. MPL-2.0 is a file-level weak copyleft: linking to the
unmodified published crates from an MIT project is permitted; modifications to
Symphonia source files would require sharing those files under MPL-2.0.
Homepage: https://github.com/pdeljanov/Symphonia
License text: https://www.mozilla.org/en-US/MPL/2.0/
Streaming Sortformer v2 (opt-in `sortformer` feature)
-----------------------------------------------------
When the optional Sortformer adapter downloads weights, they come from the
community ONNX export of NVIDIA Streaming Sortformer 4spk v2
(`nvidia/diar_streaming_sortformer_4spk-v2`), mirrored at
`cgus/diar_streaming_sortformer_4spk-v2-onnx` (export by Altunenes for
parakeet-rs).
Licensed under CC-BY-4.0 (https://creativecommons.org/licenses/by/4.0/).
Attribution: NVIDIA Corporation (Sortformer v2).
Weights are never bundled with polyvoice. Redistributors who ship the ONNX
file must provide the CC-BY-4.0 attribution above. Sortformer v2.1 (NVIDIA
Open Model License) is not used.