Licenses & Attribution
Last updated
is built on open models, open datasets and open-source software. We're grateful to the people who made them. The credits below are required by their licenses and given gladly.
Speech model
Speech is generated with Pocket TTS by Kyutai (huggingface.co/kyutai/pocket-tts). The model weights are licensed under Creative Commons Attribution 4.0 International (CC BY 4.0), and the Pocket TTS source code is licensed under the MIT License.
Changes: we run the model as a hosted service. Input text is normalized before synthesis, and the model's output is post-processed by resampling, speed adjustment and audio encoding. The model's use restrictions, including the ban on unauthorized voice cloning, impersonation and deception, are part of our Acceptable Use Policy.
Stock voices
Each stock voice is a voice profile computed from a recording in the dataset below. "Changes" for every voice: the recording was converted into a numeric voice embedding, and new speech is synthesized from that embedding.
| Voice ids | Source | License |
|---|---|---|
alba | Recording by voice actor Alba MacKenna, distributed by Kyutai | CC BY 4.0 |
marius, javert | Kyutai Voice Donation Project | CC0 1.0 |
anna, azelma, charles, eponine, eve, fantine, george, jane, mary, michael, paul, vera | CSTR VCTK Corpus, Centre for Speech Technology Research, University of Edinburgh (Yamagishi, Veaux & MacDonald) | CC BY 4.0 |
bill_boerst, caro_davy, peter_yearsley, stuart_bell | Voice-Zero collection of LibriVox volunteer recordings (Bill Boerst, Caro Davy, Peter Yearsley, Stuart Bell) | CC0 1.0 / public domain |
Voice names in our catalogue are product labels and do not identify the original speakers, except where a speaker is credited above.
Your audio
You may use audio you generate with stock voices commercially, subject to our Terms and Acceptable Use Policy. We don't require you to credit us. Audio generated with a cloned voice is your responsibility under the Voice Cloning Terms.
Open-source software
- FFmpeg: audio decoding and encoding, run as a separate server-side program. LGPL/GPL.
- PyTorch (BSD-3-Clause), FastAPI (MIT), NumPy (BSD-3-Clause): the synthesis server.
- Firebase JavaScript SDK: website sign-in. Apache License 2.0.
- Inter and JetBrains Mono typefaces: SIL Open Font License 1.1.
Spotted a missing or incorrect credit? Email and we'll fix it.