Skip to main content

start to finish

How making an AI cover actually works

Two jobs, in order. First Voxmin learns your voice, once. Then it uses that voice on any song you choose, as many times as you like.

Short answer

Training happens once and takes a while. Covering happens in minutes and can be repeated for free-form experimentation. People usually expect the opposite, which is why the first attempt feels slow and every one after it feels instant.

  1. You give it recordings

    About ten minutes of one person singing, clean, with nothing playing behind them. More is not automatically better; clean is better.

  2. It learns, once

    This runs in the background and you can close the tab. You are told when it is done. You never have to do this step again for that voice.

  3. You choose a source song

    This is the performance being covered. Voxmin separates the singing from the music so it can replace one without disturbing the other.

  4. It sings and hands you the file

    A few minutes later you have an audio file. If it is not right, change the pitch setting and run it again — the training is already done.

Questions people actually ask

How long does training take?
Long enough that you should start it and go do something else. It runs on our machines, not yours, so closing the browser is fine.
How long does one cover take?
A few minutes for a normal-length song.
What is the pitch setting for?
If the original singer sings much higher or lower than you comfortably do, the result can sound strained. Shifting the pitch up or down in semitone steps moves the song into your range.

Hear a song in your own voice

Train your voice once, then cover anything you have the right to use. Pay per song, and the credits do not expire.