How to remove vocals from a song (karaoke) with AI in your browser

For years, 'removing vocals' meant subtracting the left channel from the right and hoping: it only worked if the voice was dead centre, and it took the bass and drums with it. Today AI models separate sources: they learn what a voice sounds like and pull it out of the mix.

Which model we use

Our Vocal Remover uses Demucs, an open-source model trained to separate vocals, drums, bass and other instruments. It downloads once (about 180 MB) and then runs on your device: the song is never uploaded.

Step by step

1. Open the Vocal Remover and pick the song (MP3, WAV, FLAC, M4A or OGG).
2. Click separate. The first time, the model downloads; then it takes one to several minutes depending on your device.
3. Listen to both tracks: instrumental (for karaoke) and vocals (acapella).
4. Download the ones you need.

What if I want more tracks?

The Stem Splitter uses the same model but delivers four tracks: vocals, drums, bass and instruments. Great for practising an instrument, remixing or studying an arrangement.

What to expect

• Backing vocals and reverb: vocal echoes and heavily mixed backing vocals may linger faintly in the instrumental.
• Voice-like instruments: a sax or very 'vocal' synth may bleed into the vocal track.
• Old or mono recordings: separate less cleanly than a modern stereo production.

Is it legal?

Separating tracks for personal use (singing at home, practising, studying) is common. Publishing an instrumental or acapella of a copyrighted song without permission is not. If you'll publish, use your own or licensed music.