# Third-Party Notices Last reviewed: 2026-08-17 ## Scope The project-authored software is licensed under the Business Source License 1.1 in `LICENSE`, with a four-year change to the MIT License for each version. That license does not replace or override the licenses of third-party software, model weights, services, media, data, or other materials used by or stored in this repository. Those materials remain subject to their own terms. This file records the third-party materials that are material to the project's current source and default runtime. It must accompany source and binary distributions of the project. A distributor must also retain any copyright, license, attribution, source-offer, or `NOTICE` files supplied with the exact third-party artifacts included in its distribution. ## Important non-commercial model restrictions The default pipeline uses the DiariZen and DialogueSidon model weights. Both are offered under Creative Commons Attribution-NonCommercial 4.0 International (`CC-BY-NC-4.0`). The project's Business Source License does not grant commercial-use rights for those weights. A deployment using either model must comply with the non-commercial restriction or obtain separate permission from the applicable rights holder. Replacing a restricted model does not by itself establish that the resulting deployment is compliant. The replacement and all other included dependencies, models, inputs, and outputs must be reviewed under their own terms. ## Source-code components ### DiariZen - Component: DiariZen speaker-diarization toolkit - Upstream: - Revision: `844f5555b0a98acd0931511fc641a8c5b8ba92c7` - Used by: `tasks/diarize-audio-part` - Distribution: retrieved from Git at installation time; not vendored in this source repository - License: MIT - Copyright: Copyright (c) 2024 BUT Speech@FIT - Project changes: no upstream source files are modified in this repository; project-authored adapter code calls the upstream package ### pyannote.audio fork distributed with DiariZen - Component: the `pyannote-audio` subdirectory of DiariZen - Upstream: - Revision: `844f5555b0a98acd0931511fc641a8c5b8ba92c7` - Used by: `tasks/diarize-audio-part` - Distribution: retrieved from Git at installation time; not vendored in this source repository - License: MIT - Copyright: Copyright (c) 2020 CNRS - Project changes: no upstream source files are modified in this repository ### pyannote.audio - Component: pyannote.audio - Upstream: - Resolved version: `4.0.7` in `tasks/split-raw-audio-into-parts/uv.lock` - Used by: `tasks/split-raw-audio-into-parts` - Distribution: retrieved from the Python package index at installation time; not vendored in this source repository - License: MIT - Copyright: Copyright (c) 2020 CNRS - Project changes: no upstream source files are modified in this repository The following MIT terms apply to the MIT-licensed components and copyright notices in this section: > Permission is hereby granted, free of charge, to any person obtaining a copy > of this software and associated documentation files (the "Software"), to deal > in the Software without restriction, including without limitation the rights > to use, copy, modify, merge, publish, distribute, sublicense, and/or sell > copies of the Software, and to permit persons to whom the Software is > furnished to do so, subject to the following conditions: > > The above copyright notice and this permission notice shall be included in all > copies or substantial portions of the Software. > > THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR > IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, > FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE > AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER > LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, > OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE > SOFTWARE. ## Model components ### DiariZen WavLM Large S80 MD v2 - Model: `BUT-FIT/diarizen-wavlm-large-s80-md-v2` - Upstream: - Upstream credit: BUT Speech@FIT and the DiariZen authors - Used by: `tasks/diarize-audio-part` - Distribution: downloaded at runtime; weights are not stored in this source repository - License: Creative Commons Attribution-NonCommercial 4.0 International (`CC-BY-NC-4.0`) - License text: - Project changes: the project downloads and runs the upstream weights without modifying the weight files - Restriction: non-commercial use only unless separate permission is obtained The current configuration does not pin a model-repository revision. The upstream revision reviewed for this notice was `f27b9ffbedcf422856d104ecee9b94be37ea578e`. A distributor must verify the license and notices of the actual revision it downloads. ### DialogueSidon - Model: `sarulab-speech/DialogueSidon` - Upstream: - Revision: `d43d7478402a5527136c6733c3f4359c37b312ab` - Upstream credit: Wataru Nakata, Yuki Saito, Kazuki Yamauchi, Emiru Tsunoo, and Hiroshi Saruwatari - Used by: `tasks/separate-chunk` - Distribution: downloaded at runtime; weights are not stored in this source repository - License: Creative Commons Attribution-NonCommercial 4.0 International (`CC-BY-NC-4.0`) - License text: - Project changes: the project downloads and runs the upstream weights without modifying the weight files - Restriction: non-commercial use only unless separate permission is obtained ### NVIDIA Parakeet TDT 0.6B v3 - Model: `nvidia/parakeet-tdt-0.6b-v3` - Upstream: - Revision: `541d1f99c6b0c3cd0b11a95167540bb8edefd82b` - Upstream credit: NVIDIA - Used by: `tasks/transcribe-chunk` - Distribution: downloaded at runtime; weights are not stored in this source repository - License: Creative Commons Attribution 4.0 International (`CC-BY-4.0`) - License text: - Project changes: the project downloads and runs the upstream weights without modifying the weight files ### Microsoft WavLM Base Plus for Speaker Verification - Model: `microsoft/wavlm-base-plus-sv` - Upstream: - Revision: `feb593a6c23c1cc3d9510425c29b0a14d2b07b1e` - Upstream credit: Microsoft and the WavLM authors listed in the upstream model card - Used by: `tasks/separate-chunk` - Distribution: downloaded at runtime; weights are not stored in this source repository - License identified by the upstream model card: Creative Commons Attribution-ShareAlike 3.0 Unported (`CC-BY-SA-3.0`) - License text: - Project changes: the project downloads and runs the upstream weights without modifying the weight files ### NISQA v2.0 - Model: NISQA v2.0 `nisqa.tar` - Upstream: - Revision: `fe84f0f252abec382b24367d5b22498a7ce34dbb` - Used by: `tools/score-completed-chunks` - Code license: MIT - Model-weights license: Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (`CC-BY-NC-SA-4.0`) - License text: - Distribution: downloaded at runtime after SHA-256 verification; weights are not stored in this source repository ### Microsoft DNSMOS - Models: DNSMOS P.835 `sig_bak_ovr.onnx` and P.808 `model_v8.onnx` - Upstream: - Revision: `591184a9fcb2cbdec02520fed81a32bbbf9d73ff` - Used by: `tools/score-completed-chunks` - Copyright and license: Microsoft Corporation; see the upstream DNS Challenge repository license and notices - Distribution: downloaded at runtime after SHA-256 verification; model files are not stored in this source repository ### pyannote Segmentation 3.0 - Model: `pyannote/segmentation-3.0` - Upstream: - Upstream credit: pyannote and CNRS - Copyright: Copyright (c) 2023 CNRS - Used by: `tasks/split-raw-audio-into-parts` - Distribution: downloaded at runtime; weights are not stored in this source repository - License: MIT - Project changes: the project downloads and runs the upstream weights without modifying the weight files - Access condition: the upstream repository is gated and requires an authenticated user to accept its access conditions The current configuration does not pin a model-repository revision. The upstream revision reviewed for this notice was `e66f3d3b9eb0873085418a7b813d3b369bf160bb`. A distributor must verify the license and notices of the actual revision it downloads. The MIT terms quoted in the source-code section above apply to this model's MIT notice. ## Python and system dependencies The Python dependencies are declared in each component's `pyproject.toml` and resolved in the adjacent `uv.lock`. These packages and their transitive dependencies are not relicensed under the project's Business Source License. The source repository does not vendor their package contents. An executable, wheel, container image, virtual-machine image, or other binary distribution may include those packages. Its distributor must produce a license inventory from the exact installed distributions and include all license texts, copyright notices, attribution notices, and source-code offers required by those versions. The lock files are the authoritative version inventory for each independently deployable project; this notice is not a substitute for a binary-distribution license bundle. FFmpeg is an external runtime requirement of the ingest service and is not included in this source repository. FFmpeg licensing depends on how a particular build is configured. Anyone distributing an FFmpeg build with this project must comply with that build's applicable LGPL, GPL, and component license terms. PostgreSQL, Redis, and S3-compatible services are also external runtime services and are not distributed by this source repository. ## Bundled NaturalVoices speech/music artifacts The following model artifacts are stored directly in the source repository: - `tasks/quality-filter-audio-part/src/voice_pipeline_quality_filter_audio_part/music_artifacts/weights.28-0.13exp1_blstm.hdf5` - `tasks/quality-filter-audio-part/src/voice_pipeline_quality_filter_audio_part/music_artifacts/mean_gtzan_esc-50_muspeak_musan.npy` - `tasks/quality-filter-audio-part/src/voice_pipeline_quality_filter_audio_part/music_artifacts/std_gtzan_esc-50_muspeak_musan.npy` Provenance and license: - Component: Speech and Music Detection artifacts distributed by NaturalVoices - NaturalVoices upstream: - NaturalVoices revision: `2d816822cc2b45438a8ecc949602f3572aade44b` - Upstream path: `pipeline_code/music_noise/speech-music-detection/checkpoint/` - Original component credit: Quentin Lemaire - Copyright: Copyright (c) 2018 Quentin Lemaire - License: MIT, as stated in the specific `pipeline_code/music_noise/speech-music-detection/LICENSE` file - License source: - Used by: `tasks/quality-filter-audio-part` - Distribution: the three files are bundled in this source repository - Project changes: none to the three artifact files; their SHA-256 digests exactly match the files at the NaturalVoices revision above The MIT terms quoted in the source-code section above apply to these artifacts and must be retained together with the copyright notice. NaturalVoices describes the artifacts as part of its speech/music detection pipeline. The associated NaturalVoices publication is "Towards Naturalistic Voice Conversion: NaturalVoices Dataset with an Automatic Processing Pipeline" by A. N. Salman, Z. Du, S. S. Chandra, I. R. Ulgen, C. Busso, and B. Sisman, Interspeech 2024. ## No endorsement The names of third-party projects, authors, and rights holders are provided only for attribution. Their inclusion does not imply that they endorse AveraLabs or this project.