AI voice (RVC)
Conversion into a trained target voice — the voice lives in the model, not in the program.
Warning: The AI voice pack is not part of the app. It is a separate
download (PyTorch, 2–3 GB) built with voice/build.sh. Without it everything
else works exactly as normal.
Models
- Add a
.pthmodel and, if you have it, the.indexfile with the same name. It makes the result considerably closer. - Ready-made models can be found on weights.gg or Hugging Face; your own are trained with the RVC WebUI.
Settings
| Value | Meaning |
|---|---|
| Pitch | Semitones before conversion (m→f usually +12, f→m usually −12) |
| Similarity | How strongly the model's timbre pulls |
| Quality | Fast (~0.4 s) · Medium (~0.7 s) · Best (~1.2 s) |
Note: For relaxed talking the character mode is the better choice — it has almost no delay.
Rights
A voice model reproduces a real person. Only use voices you have permission for — your own, or one that has been explicitly released.