Lip Service Copyright (C) 2026 Giles Hall Lip Service, its engine and its trained models, is licensed under the GNU Affero General Public License, version 3 only (LICENSE). A commercial license is available from the copyright holder for use the AGPL does not suit. Two things inside the model files keep their own licenses: the Vocos weights inside fast/melvoc.lsw (MIT, below) and the CMU Pronouncing Dictionary's entries inside lexicon.lsw (below). The engine has no third-party crates and no vendored source. Compiled, it includes the Rust standard library (Copyright The Rust Project Developers, under the MIT license or the Apache License, Version 2.0). The model files, and what each was made from -------------------------------------------- The models are not in the repository. The package fetches them, and they are published with a copy of this file. base.lsw the shared speech model LibriTTS-R, from an LJ Speech start hifi/vox.lsw hifi's route to sound LibriTTS-R and VocalSet fast/melvoc.lsw fast's route for speech decoder: LibriTTS-R, from the speech model; Vocos: unmodified weights, MIT (below) fast/vocoder.lsw fast's route for singing LibriTTS-R, from an LJ Speech start voices/*.lsw the sixteen voices, and LibriTTS-R hifi/voices/*.lsw each one's part for hifi fast/voices/*.lsw and for fast breaks.lsw pause model LibriTTS-R, from the speech model; taught in part by BERT g2p.lsw letter-to-sound model CMU Pronouncing Dictionary lexicon.lsw pronunciation lexicon CMU Pronouncing Dictionary, plus Lip Service's own entries motion/straight.lsw, sung pitch motion measured from VocalSet motion/vibrato.lsw, motion/wide.lsw hifi/vox.lsw (sha256 cb78b1bd1317a9a2...) was trained on read speech from LibriTTS-R and singing from VocalSet. The engine's rule for an emphasized word was measured on LibriTTS-R's readers with the Crowdsourced LibriTTS Speech Prominence Annotations (Max Morrison, 2023, doi:10.5281/zenodo.10402793), used under CC BY 4.0. The annotations themselves are not distributed. The website's choir also plays recorded breaths: short excerpts of VocalSet recordings (CC BY 4.0, below), served by the site in breaths.bin. They are not in the repository or the package. LibriTTS-R ---------- The speech model, the voices, both routes to sound and the pause model were trained on LibriTTS-R. Yuma Koizumi, Heiga Zen, Shigeki Karita, Yifan Ding, Kohei Yatabe, Nobuyuki Morioka, Michiel Bacchiani, Yu Zhang, Wei Han, and Ankur Bapna. "LibriTTS-R: A Restored Multi-Speaker Text-to-Speech Corpus." Interspeech 2023. https://www.openslr.org/141/ The LibriTTS-R corpus is made available by Google LLC under a Creative Commons Attribution 4.0 International License: https://creativecommons.org/licenses/by/4.0/ LibriTTS-R is derived from LibriTTS (Zen et al., Interspeech 2019), which is derived from LibriSpeech (Panayotov et al., ICASSP 2015), which is drawn from LibriVox audiobook recordings that their readers released into the public domain. The model files are models trained on the recordings; no LibriTTS-R recording is distributed. Each voice has an invented first name. No reader made, endorses or is associated with Lip Service. Each voice is tuned to one reader of LibriTTS-R, and the package's `voices` command prints that reader's number in the corpus. LJ Speech --------- The speech model and the vocoder started from models trained on the LJ Speech Dataset; the mel decoder and the pause model were trained from the speech model. Keith Ito and Linda Johnson. "The LJ Speech Dataset." 2017. https://keithito.com/LJ-Speech-Dataset/ The LJ Speech Dataset is in the public domain. VocalSet -------- The sung pitch motions (motion/straight.lsw, motion/vibrato.lsw, motion/wide.lsw) were measured from VocalSet's recordings, and hifi's route to sound (hifi/vox.lsw) was trained on them beside LibriTTS-R. No VocalSet recording is in the model files. The website's choir breaths (breaths.bin, served by the site only) are 76 short excerpts of VocalSet recordings: the breaths singers took between notes. breaths.json, served beside it, names the recording and the time span each one came from. Changes made: each was cut to the breath alone, resampled to 24 kHz, set to a common level and raised (5.6 times that level), and stored as 16-bit samples. The excerpts remain under VocalSet's license. Julia Wilkins, Prem Seetharaman, Alison Wahl, and Bryan Pardo. "VocalSet: A Singing Voice Dataset." ISMIR 2018. Version 1.2. https://doi.org/10.5281/zenodo.1442513 VocalSet is licensed under a Creative Commons Attribution 4.0 International License: https://creativecommons.org/licenses/by/4.0/ Vocos ----- fast/melvoc.lsw contains the weights of Vocos (charactr/vocos-mel-24khz), unchanged except that its pointwise layers are stored as 8-bit integers. Vocos was trained by its authors on LibriTTS. The engine's Vocos code is Lip Service's own implementation. These weights remain under the MIT License below. Hubert Siuzdak. "Vocos: Closing the gap between time-domain and Fourier-based neural vocoders for high-quality audio synthesis." arXiv:2306.00814, 2023. https://github.com/gemelo-ai/vocos MIT License Copyright (c) 2023 Charactr Inc. 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. BERT ---- The pause model (breaks.lsw) was trained in part to match the predictions of BERT (bert-base-uncased, released by Google under the Apache License 2.0). No part of BERT is distributed. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. "BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding." NAACL 2019. arXiv:1810.04805. https://huggingface.co/google-bert/bert-base-uncased CMU Pronouncing Dictionary -------------------------- The lexicon is the CMU Pronouncing Dictionary, packed, and the letter-to-sound model was trained on it. The lexicon also holds entries written for Lip Service, which are not part of the CMU dictionary and are under the AGPL with the rest. Copyright (C) 1993-2015 Carnegie Mellon University. All rights reserved. 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.