How to make an AI cover with your own RVC model

How it works

1. Drop the song. Separate the stems first is already ticked on this page, because a cover starts from a full song. Untick it if your file is already a clean solo vocal.
2. Choose a voice. Use a voice you trained on NiceVois, drop in an RVC .pth or model ZIP you already have, or open the NiceVois voice library tab and pick a credited voice.
3. Set the pitch and convert. You get the mixed cover, plus the converted lead vocal, the instrumental and the backing vocals as separate files, so you can mix it yourself if you prefer.

Get the pitch right

This is the setting that decides whether the result sounds like a voice or a chipmunk. Pitch is measured in semitones, and twelve semitones is one octave.

When the result is poor

• Robotic or garbled: the model heard the whole mix. Tick Separate the stems first and convert again.
• Wrong octave: change the pitch, per the table above.
• The voice sings against itself: harmonies were converted along with the lead. With stems separated, only the lead is converted; why that matters.
• Doesn't resemble the voice: usually the training audio, not the conversion. See dataset preparation.
• Breathy or unstable on held notes: the model may be undertrained. See the epoch guide.

Doing it by hand in another RVC tool

If you run RVC yourself, the converter above does these steps for you, but the sequence is the same:

1. Separate the vocal. Split the song into an isolated vocal and an instrumental, and keep both. If the song has harmonies or doubles, split the lead from the backing vocals too.
2. Convert the vocal. Load your .pth and its matching .index. A .pth without its index sounds thinner and more artefact-heavy. The index ratio pulls the result toward the training data when high and tracks the input more literally when low; start mid-range.
3. Set the pitch using the table above.
4. Mix it back. Line the converted vocal up over the instrumental, then level it, add a little compression, and add reverb to seat it. A converted vocal almost always arrives drier than the original.

Common questions

Why does my AI cover sound bad?

Usually because the model was fed a finished mix instead of an isolated vocal. On this page, Separate the stems first handles that: the lead is split out, converted on its own, and mixed back with the music. If you run RVC yourself, separate the track first and convert only the vocal.

What are the steps to make an AI cover with an RVC model?

Separate the song into a vocal and an instrumental, convert the vocal with your voice model, set the pitch so the result sits in the voice's natural range, then mix the converted vocal back over the instrumental. The converter on this page does the separating, converting and mixing from one full song.

Do I need my own voice model?

No. You can use a voice you trained on NiceVois, drop in an RVC .pth or model ZIP you already have, or choose a credited voice from the NiceVois voice library.

Can I publish an AI cover I made?

Only if you have the rights. A trained voice model does not grant permission to use someone's voice, and the underlying song has its own separate rights.