Weights AI

Train an AI voice model

Upload a few minutes of a voice, and use it on any song once it's trained.

  1. 1Upload recordings of the voice

    • Sing or speak a cappella: no music, no backing tracks.
    • Record in a quiet room, close to the mic, without reverb or effects.
    • 10 minutes or more works best (at least 5); more variety (high, low, soft, loud) helps.
    • Only your own voice, or a voice you have permission to use.

  2. 2Name it

    Who can use it

  3. 3Generate

    Training costs 100 credits and takes about 15 to 30 minutes.

    Sign in with Google or email to get 20 credits a day, about 2 covers.

How voice training works

  1. 1

    Upload recordings

    10 minutes or more of clean, a cappella singing or speech.

  2. 2

    Train

    We train an RVC v2 model on a cloud GPU in about 15 to 30 minutes.

  3. 3

    Make covers

    Your voice appears in the cover tool, ready for any song.

Questions

Whose voice can I train?

Your own, or a voice you have permission to use. Models are private unless you choose to share them.

What makes a good recording?

A quiet room, no music or effects, and a mix of high and low, soft and loud singing. More clean audio beats more total audio.

What does it cost?

Training uses 100 credits. If training fails, the credits are refunded automatically.

Can I use RVC models I already have?

Model import (.pth and .index files) is coming next; for now, train a new model here.