Legal
Licences and credits
Vozeru is a thin Rust shell around other people's research. This is who, and under what terms. Licence names are as the authors published them; the linked repositories carry the text.
Models that ship through the app
| Model | Does | Licence |
|---|---|---|
| Kokoro-82M | Text to speech. The voice packs on the Text tab are its voices. | Apache-2.0 |
| misaki | Text to phonemes, the front end Kokoro was trained on. Its lexicons and tagger are bundled in the app. | Apache-2.0 |
| ContentVec | Speech content encoder for voice conversion (the 768-dimension, layer 12 export used by RVC). | MIT |
| RMVPE | Pitch estimation for voice conversion. | MIT |
| RVC | The voice conversion architecture every catalogue voice is a model for. Vozeru runs its generator on the phone. | MIT |
| UVR-MDX-NET-Voc_FT | Vocal and instrumental separation for music mode, from the Ultimate Vocal Remover project. | MIT |
The models are downloaded from a Vozeru-maintained mirror on Hugging Face, unmodified apart from a retiming of the separator graph that lets it run in a phone's memory.
Catalogue voices
Each voice on the Voices tab is an RVC model published by someone else on Hugging Face, under the licence they chose, or none. Vozeru indexes them, does not host them, and removes an entry on request. The person who trained a voice is credited by name on its entry where the source gives one.
Software the app is made of
| Component | Does | Licence |
|---|---|---|
| kokoro-rs, rvc-rs, uvr-rs, misaki-rs | The four engines, Rust ports of the pipelines above. | See each repository |
| ONNX Runtime via ort | Runs every model. | MIT; Apache-2.0 or MIT |
| Dioxus | The user interface. | MIT or Apache-2.0 |
| Symphonia | Decodes the audio you bring in. | MPL-2.0 |
| hound | Writes the WAV files that come out. | Apache-2.0 |
| rusqlite and SQLite | The catalogue search. | MIT; public domain |
| ureq and rustls | The downloads. | MIT or Apache-2.0 |
| Tokio | Timers. | MIT |
This website
| Thing | Licence |
|---|---|
| Metropolis typeface | Unlicense |
| Katakana rendered from Noto Sans CJK | SIL Open Font License 1.1 |
| The mark, the wordmark, Vo and the rest of the art | © Vozeru. Not for reuse outside the app and this site. |
Trademarks
Google Play and the Google Play logo are trademarks of Google LLC. Android is a trademark of Google LLC. Hugging Face is a trademark of Hugging Face, Inc. None of them endorse Vozeru.