Skip to content

Karaoke maker: remove the vocals and sing in your key

Upload a song, the AI takes out the lead vocal, and you sing over the band in a key that suits your voice.

MP3, WAV, FLAC, M4A or OGG, up to 10 minutes long

  • ~1-3 min processing
  • Up to 64 MB
  • MP3, WAV, FLAC, M4A, OGG
  • Deleted 24 h after processing without an account, 14 days with one (90 on Pro)
Original key

Tap a lane to mute it or bring it back.

Switch between the original and the karaoke version, then move the key and the tempo like on the practice screen.

Demo song: “Ślad” by 3Biegun, separated with Demucs, the model behind this tool, and played here with the band's permission.

How to make a karaoke track from a song

  1. 1

    Upload the song

    MP3, WAV, FLAC, M4A or OGG, up to 64 MB. Without an account it can be 7 minutes long, with a free account 10.

  2. 2

    Let the AI take the voice out

    The separator splits the recording into vocals, drums, bass and the rest. Most songs are ready in one to three minutes.

  3. 3

    Sing over the band

    The practice screen opens with the vocals muted. Change the key, slow it down, loop the chorus, and download the backing track with a free account.

Sing it in your key, not the original singer's

A song recorded for someone else's voice often sits too high or too low for yours. On the practice screen you move the whole backing track up or down, up to six semitones, and the band still sounds like a band.

  • Find your key

    Choose your voice type and the lowest and highest notes you sing comfortably. The screen compares that with the vocal line and suggests how far to move the song.

  • Keep the singer as a guide

    The original vocal can play at full level, at about 30% as a guide, or not at all. Start with the guide and turn it off once you know the melody.

  • Learn it slowly

    Take a fast verse down to 75% or 50% without changing the key, then work back up to full speed.

  • Take it with you

    With a free account, download the backing track in the key and tempo you set, for practice at home or a rehearsal.

What you can do once the song is split

Every separation opens on the practice screen, with one lane per stem. Listening and practising there need no account.

  • Change the key

    Move the whole song up or down, a semitone at a time, up to six either way.

  • Find your key

    Pick your voice type and your comfortable range, and the screen suggests how far to move the song.

  • Keep a guide vocal

    Leave the original singer at about 30% while you learn the tune, then switch it off.

  • Slow it down or speed it up

    Anywhere from 40% to 150%, and the key stays where it is.

  • Loop the hard part

    Mark a section, give it a name and play it until it's clean.

  • Download in your key

    With a free account, download the mix in the key and tempo you set, or any stem as the separator made it.

What's free

Separation runs on our own GPU server and is counted in minutes of audio. A four-minute song uses four minutes, whichever split you pick.

Without an account

One song a day, up to 7 minutes long. Listen, loop and change the key and tempo. Results stay for 24 hours.

With a free account

15 minutes of separation a month, songs up to 10 minutes, and MP3 downloads. Results stay for 14 days.

When you need more

Minute packs never expire, and a separation that fails gives its minutes back. Pro adds WAV files, priority in the queue and 90 days of storage.

See minute packs and Pro

Frequently asked questions

Can I change the key of the karaoke track?
Yes. On the practice screen the key moves a semitone at a time, up to six up or six down, and the tempo goes from 40% to 150%. The whole backing moves together, so the instruments stay in tune with each other. With a free account you can download it in the new key.
Why can I still hear some of the vocals?
Reverb tails and dense harmonies sit close to the instruments, so faint traces can stay in a busy chorus. A WAV, FLAC or 320 kbps MP3 gives the model more to work with than a low-bitrate copy.
Are the backing vocals removed as well?
Yes. The model puts the lead vocal, the harmonies and the doubles in one vocal stem, so muting it takes all of them out. It can't keep the backing vocals and drop only the lead.
Can I keep the original singer quietly in the background?
Yes. The guide vocal has three settings: full, guide (about 30% of its level) and off. Start with the guide, and switch it off once you know the tune.
Can I paste a YouTube link?
No. The tool works on a file you upload: MP3, WAV, FLAC, M4A or OGG, up to 64 MB. Use a recording you have the right to use, such as your own or one you have permission for.
Can I use the backing track in karaoke software?
Yes. Download it as an MP3 with a free account (WAV on Pro) and it plays in any karaoke app or media player. For lyrics on screen you still need a lyric file or an app that adds them.
Is it free?
Listening and practising are free. Without an account you can separate one song a day, up to 7 minutes long. A free account gives you 15 minutes of separation a month and MP3 downloads. If you need more, minute packs never expire, and Pro adds WAV files.
How long does separation take?
Most three- or four-minute songs are ready in one to three minutes. It depends on the length of the file and on how busy the server is. A song can be up to 10 minutes long, or 7 without an account.
Is my file uploaded somewhere permanent?
Files are processed on our own server. Without an account they're deleted 24 hours after the job completes, on a free account after 14 days and on a paid one after 90. We don't keep your audio beyond that. Files uploaded from 1 October 2026 may be used to train our own AI models while we keep them (see the Terms of Service).

Continue your workflow

Once your stems are separated, take them straight into another tool.