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.
