# Third-party components

SmollerTTS original code and acoustic weights: MIT, Copyright 2026 Dun Li Chan.
The pretrained HiFi-GAN LJ_V2 vocoder and adapted generator definition retain
Jungil Kong's MIT license in [HIFIGAN-LICENSE.txt](licenses/HIFIGAN-LICENSE.txt).
Source: https://github.com/jik876/hifi-gan . Original generator SHA-256:
3fac378c5918fb2c102733f21eeaa8e9a4ca6cda24dbfddc55bbb947c78d562f.

ONNX Runtime Web 1.31.1: Microsoft, MIT; [license](licenses/ONNXRUNTIME-LICENSE.txt).
https://github.com/microsoft/onnxruntime

phonemize 2.0.1 English rules/data: Copyright 2025 Hans, MIT; [license](licenses/PHONEMIZE-LICENSE.txt).
https://github.com/hans00/phonemize

to-words 7.5.0: Munjal Dhamecha, MIT; [license](licenses/TO-WORDS-LICENSE.txt).
https://github.com/mastermunj/to-words

The English pronunciation data in en.lex is adapted from the MFA English (US)
dictionary by Michael McAuliffe and Morgan Sonderegger, under CC BY 4.0.
Source: https://mfa-models.readthedocs.io/ ; license:
https://creativecommons.org/licenses/by/4.0/ . Changes: selected supported words
and pronunciations, mapped phone/stress IDs, and packed records for browser use.
MFA does not endorse this model. The exact delivered binary is pinned in
[model/assets.lock.json](model/assets.lock.json).

LJSpeech recordings used for acoustic training are described by their provider
as public domain in the US: https://keithito.com/LJ-Speech-Dataset/ . Original
training recordings are not included. Samples are outputs generated by this model.

Geist: Copyright 2024 The Geist Project Authors, SIL Open Font License 1.1;
[license](licenses/Geist-LICENSE.txt). https://github.com/vercel/geist-font

GitHub and Hugging Face marks identify links to those services. They remain
their respective owners' marks and do not imply sponsorship or endorsement.
https://github.com/logos · https://huggingface.co/brand

Published benchmark data is attributed to
[tts-bench](https://github.com/5uck1ess/tts-bench/tree/1dd3a81a50dd21fb2f4bce89df27eada996ba57f).
Per-model links, source revision and measurement details are included with the data.
