Extract vocals. Completely free.

Yes, completely. Every tool is free, and there's no paid version, account, watermark or daily limit.

No. The model runs on your own device, so your recording never leaves it. Once the model has downloaded, this works offline too.

No. There is no account, no paid version and no monthly quota, because your device does the work rather than a server someone has to pay for.

It uses HTDemucs, the model behind many paid stem-splitting services. Finished, mixed songs work best. Expect a very clean instrumental, and vocals with faint traces of the track on dense or heavily reverberant mixes.

On a computer with a graphics card, usually under a minute. Without one it still works but can take many minutes per song, so try a short clip first.

Vocals and an instrumental, or four stems: vocals, drums, bass and everything else. You can listen to them together, mute or solo each one, then save them separately.

WAV, MP3, FLAC, M4A or OGG. You can drop nearly any audio format, or a video to use its sound.

The separation is yours to use, but the rights in the song still belong to whoever owns them. Personal practice and karaoke are one thing; publishing someone else's music is another.