Conversion, not synthesis
The model keeps your performance and swaps the timbre. Phrasing, pauses and pitch movement are yours. It Is why converted singing still sounds like a person and not a machine.
What to upload
A clean recording with one voice. Background music and room noise confuse the conversion, so run the noise remover first if the source is rough.
Consent is not optional here
Converting your own voice is fine. Converting someone else's to make them appear to say something is impersonation, and it breaks the terms. The consent box is enforced.
FAQ
Does it work on singing?
Yes. Melody and timing are preserved, only the timbre changes.
How is this different from voice cloning?
Cloning builds a voice from a sample and reads new text. Conversion takes an existing recording and changes how it sounds.
What files work?
MP3, WAV and M4A up to 30 MB.