Credits and licences

Last updated: October 30, 2026

KUKU Labs is built on the work of others, who shared it openly. This page says who made what, under which licence, and what we changed. Crediting someone does not mean they endorse KUKU Labs.

1. Voice cloning and the classic voices: Kyutai Pocket TTS

Voice cloning and the classic voices use Pocket TTS, a text-to-speech model made by Kyutai. The models are © Kyutai and are shared under the Creative Commons Attribution 4.0 International licence (CC BY 4.0). Source: huggingface.co/kyutai/pocket-tts.

What we changed: the model files used on this website are converted copies. They were converted to the ONNX format and compressed (8-bit quantisation) with community tools, so that they can run inside your browser. We have not retrained the models.

The conversions and the tools they come from are KevinAHM’s pocket-tts-onnx-export (Apache License 2.0). The models’ own terms of use also apply: they must not be used for voice impersonation or cloning without explicit and lawful consent, or to deceive people (see Kyutai’s prohibited-use conditions). Our Terms require the same of you.

2. The Voice Library: Phonon and Gradium’s voices

The ready-made voices in the Voice Library are generated with Phonon, made by Gradium, using the model gradium/phonon-31m-en (Apache License 2.0). The Phonon runtime (phonon-tts) is shared under the Apache License 2.0 or the MIT licence; its licence texts and notice are served with it at /assets/vendor/phonon-0.4.0/NOTICE.txt. The voice list and the sample recordings are Gradium’s gradium/phonon-voices (Apache License 2.0). We show them unchanged apart from the layout.

3. The Voice Enhancer: DeepFilterNet

The Voice Enhancer removes background noise with DeepFilterNet 3 (Hendrik Schröter and others), shared under the Apache License 2.0 or the MIT licence. Its licence texts are served with the model files at /assets/models/dfn3/.

4. Software and hosting

ONNX Runtime Web (Microsoft, MIT licence) runs the models in your browser. It is delivered by jsDelivr.
SentencePiece (Google, Apache License 2.0) turns text into the pieces the model reads.
lamejs (a JavaScript port of the LAME MP3 encoder, GNU LGPL 3.0) makes the MP3 downloads, in your browser. It is used unmodified as a separate file: /assets/vendor/lamejs-1.2.7/.
Hanken Grotesk (SIL Open Font License 1.1) is the typeface, served from this website.
Hugging Face hosts the model files and the voice samples that your browser downloads.

5. Names

Kyutai, Gradium, Hugging Face, Microsoft, Google and the other names on this page belong to their owners and are used only to say whose work this is.

6. Corrections

If we have credited something wrongly or left something out, please tell us at kukulabsonline@gmail.com and we will put it right.