| loudkit |
| Copyright 2026 LoudReader |
|
|
| This product includes software developed at LoudReader. |
|
|
| Licensed under the Apache License, Version 2.0. See LICENSE. |
|
|
| -------------------------------------------------------------------------------- |
| Third-party notices |
| -------------------------------------------------------------------------------- |
|
|
| Chatterbox TTS |
| Copyright (c) 2025 Resemble AI |
| Licensed under the MIT License. |
| https: |
|
|
| The weights distributed with this project — the `loudr` series — are derived |
| from Chatterbox and optimised for real-time synthesis. Parts of the pipeline |
| were retrained; others, among them the tokenizer and speaker encoder that |
| enrollment uses, are carried from the upstream artefact unchanged and are |
| redistributed here under the MIT terms above. |
|
|
| The architectures in `python/loudkit/models/` are reimplementations written |
| against those weights; they are not copies of the upstream source, but the |
| weights they load could not exist without it. |
|
|
| Voice profiles |
| The shipped voice profiles — twenty-eight enrollments across ten |
| languages, |
| distributed on the Hugging Face repository under voices/profiles/ — |
| are enrolled from the recordings catalogued in |
| `docs/voices/roster/provenance.json`, which records donor or speaker, |
| source, licence, consent basis, reference construction and sha256 for |
| every profile, reference and sample. None of them are Chatterbox |
| artefacts. Every source below was recorded or released for speech- |
| technology use, and each profile ships under its own terms: |
|
|
| OHF-Voice / NabuCasa voice-datasets — CC0-1.0 |
| https: |
| Voice donations recorded expressly for building TTS voices: |
| darkman, gosia (pl); joe, kathleen (en); kerstin (de); |
| pim, nathalie (nl); dave (es); paola (it); tugao (pt). |
| The kathleen reference reads prompts from the CMU ARCTIC prompt |
| list; those prompt texts come from public-domain novels and impose |
| no restriction on the recording itself. |
|
|
| Thorsten-Voice, TV-44kHz-Full — CC0-1.0 |
| https: |
| thorsten (de), under the donor's own public-domain dedication. |
| |
| NST Swedish and NST Danish speech databases — CC0-1.0 |
| Språkbanken, National Library of Norway: |
| https://www.nb.no/sprakbanken/en/resource-catalogue/oai-nb-no-sbr-17/ |
| https://huggingface.co/datasets/alexandrainst/nst-da |
| nils, selma (sv); soren, freja (da). |
| |
| Kyutai tts-voices — CC-BY-4.0 |
| https://huggingface.co/kyutai/tts-voices |
| henri, colette (fr). |
| |
| Kyutai Voice Donations — CC0-1.0 |
| https://huggingface.co/kyutai/tts-voices, under voice-donations/ |
| clara, emma, henry, lucy, miles, oliver, oscar, sophie (en). |
| Donations to the Unmute Voice Donation Project, whose terms |
| expressly cover speech synthesis, voice embeddings and commercial |
| use: |
| https://kyutai.org/legal/Terms%20of%20Use%20-%20Unmute%20Voice%20Donation%20Project%20v1.pdf |
| The names above are product aliases, not donor identities; the |
| roster records each donation's speaker id, revision and sha256. |
|
|
| CML-TTS — CC-BY-4.0 |
| https: |
| carmen (es), ines (pt). LibriVox-derived recordings; the dataset's |
| terms credit the original LibriVox readers. |
| |
| Multilingual LibriSpeech — CC-BY-4.0 |
| https://huggingface.co/datasets/facebook/multilingual_librispeech |
| dante (it). LibriVox-derived; likewise credits the readers. |
| |
| Voice encoder (`ve.safetensors`) |
| The 5.7 MB utterance voice encoder that drives voice cloning is an export |
| of the `VoiceEncoder` model from the Chatterbox T3 architecture (Resemble |
| AI, MIT — see above), produced by this project's own enrollment export |
| (this project's enrollment export tooling). It is a |
| derived work of the Chatterbox weights and is redistributed under the same |
| MIT terms. |
| |
| `docs/PROVENANCE-voice-encoder.md` records the artefact's size and sha256, |
| its tensors, the upstream repository and licence, and an exact tensor match |
| with ResembleAI/chatterbox revision |
| 5bb1f6ee58e50c3b8d408bc82a6d3740c2db6e18. The original export's checkout |
| revision was not recorded. |
| |
| MIT License text: |
| |
| 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. |
| |
| Enrollment architectures |
| `python/loudkit/models/enroll.py` reimplements four inference paths whose |
| designs come from the projects below. The weights those modules load are |
| part of the Chatterbox-derived checkpoint (MIT, above); what is owed here is |
| attribution for the architectures, all of them permissively licensed and all |
| compatible with this project's Apache-2.0 terms. |
|
|
| CosyVoice — Apache License 2.0 |
| Copyright FunAudioLLM / Alibaba Group |
| https: |
| The S3 speech tokenizer's shape (whisper-style encoder, FSQ quantiser, |
| 25 Hz token rate) and the matcha-recipe mel the flow conditions on. |
| |
| S3Tokenizer — Apache License 2.0 |
| Copyright Xingchen Song |
| https://github.com/xingchensong/S3Tokenizer |
| The standalone tokenizer extraction this port follows, including the |
| FSMN memory over the attention values. |
| |
| 3D-Speaker (CAM++) — Apache License 2.0 |
| Copyright ModelScope / Alibaba Group |
| https://github.com/modelscope/3D-Speaker |
| The CAM++ x-vector encoder: FCM front end, TDNN-dense body with |
| context-aware masking, statistics pooling. |
| |
| FunASR — MIT License |
| Copyright ModelScope / Alibaba Group |
| https://github.com/modelscope/FunASR |
| The FSMN formulation the tokenizer's attention block carries. |
|
|
| Real-Time-Voice-Cloning — MIT License |
| Modified & original work Copyright (c) 2019 Corentin Jemine |
| Original work Copyright (c) 2018 Rayhane Mama |
| Original work Copyright (c) 2019 fatchord |
| Original work Copyright (c) 2015 braindead |
| https: |
| The utterance voice encoder: 3-layer LSTM over 40-mel partials, |
| L2-normalised, which is the recipe its weights were trained against. |
|
|
| CMUdict (Carnegie Mellon Pronouncing Dictionary) |
| Copyright (C) 1993-2015 Carnegie Mellon University. All rights reserved. |
|
|
| `python/loudkit/models/data/pl_en_respell.json` is generated from CMUdict by |
| `tools/gen_pl_respell.py`, which maps ARPAbet phonemes to Polish |
| orthography. It is a derived work and it ships in the distribution, so the |
| notice below is reproduced as clause 2 requires. |
|
|
| Redistribution and use in source and binary forms, with or without |
| modification, are permitted provided that the following conditions |
| are met: |
|
|
| 1. Redistributions of source code must retain the above copyright |
| notice, this list of conditions and the following disclaimer. |
| The contents of this file are deemed to be source code. |
|
|
| 2. Redistributions in binary form must reproduce the above copyright |
| notice, this list of conditions and the following disclaimer in |
| the documentation and/or other materials provided with the |
| distribution. |
|
|
| This work was supported in part by funding from the Defense Advanced |
| Research Projects Agency, the Office of Naval Research and the |
| National Science Foundation of the United States of America, and by |
| member companies of the Carnegie Mellon Sphinx Speech Consortium. We |
| acknowledge the contributions of many volunteers to the expansion and |
| improvement of this dictionary. |
|
|
| THIS SOFTWARE IS PROVIDED BY CARNEGIE MELLON UNIVERSITY ``AS IS'' AND |
| ANY EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, |
| THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR |
| PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL CARNEGIE MELLON UNIVERSITY |
| NOR ITS EMPLOYEES BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
| SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT |
| LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
| DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
| THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
| (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
| OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
|
|
| Unicode CLDR |
| Copyright © 1991-2026 Unicode, Inc. All rights reserved. |
| Licensed under the Unicode License v3 — https: |
|
|
| `tests/data/cldr |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
