Free AI vocal remover

How to remove vocals from a song

Drop the song, wait for the split and download the instrumental. This online vocal remover works on any file you have: you can remove vocals from an MP3, a WAV or any file your browser can play, with the same result. The instrumental is everything that is not the voice: the drums, the bass and all the other instruments added together. You can listen to it before you save, and switch to the original to hear what changed, so you know the track is the one you wanted.

The model separates vocals from music, and it treats lead vocals and backing vocals as vocals, so both come out. If a song has a choir or heavy harmonies, they go as well. The vocal itself is kept in its own stem if you want it, so you never lose it by making the instrumental.

Make a karaoke track

A karaoke maker, or any instrumental maker, needs one thing: the song without the singer. Drop the track you want to sing, download the instrumental and play it in any karaoke app, a media player or your DAW, with the lyrics in front of you. It works for a party, a rehearsal or a recording of your own cover.

We make the audio only. We do not add lyrics or timing, and we do not change the key or the tempo. If a song sits too high for you, change its key in your DAW after you have the instrumental.

Instrumentals for covers, practice and teaching

Singers use an instrumental to rehearse a song without the original voice leading them, or to record a cover over the real arrangement. Teachers use it to give a student a backing track for a song they chose. Guitarists, keyboardists and drummers can go further: split the song into stems on the stem splitter page, mute the instrument they play and play it themselves with the rest of the band.

If you want a clean backing track for a live set, remove the vocals, listen to the result all the way through and fix any spot where the voice leaks with a short edit.

Why AI works better than the old phase trick

The classic way to remove a voice subtracts the left channel from the right. Anything that sits exactly in the middle of the stereo picture cancels, and the lead vocal is usually there. So are the kick, the snare and the bass, which is why those tricks leave a thin, hollow song, and why they do nothing on a mono file.

A neural network does not look at where a sound sits. It has learned what a voice sounds like, with its harmonics and consonants, and it takes that out while the rest of the song stays in place. That is also why it can separate vocals that are not in the centre.

What to expect, honestly

Most songs come out clean enough to sing over or to practice with. Some will not be perfect. You can hear a faint ghost of the voice, a trace of reverb or a breath, especially in a dense mix, with a heavily processed vocal or with instruments that sound like a voice, like a vocal chop or a synth lead. Low-bitrate MP3s make this worse, because compression blurs the detail the model relies on.

For a clean result, start from a WAV or a high-bitrate file, and listen to the instrumental in full before you rely on it. If you still hear the voice, run the song again in HQ: it is a newer model that separates more cleanly on public benchmarks.

Private, free and in beta

The work happens on your device, so your song is never uploaded and we have no server bill that would force a limit on you. There is no sign-up, no credit counter and no daily cap, just a limit of 10 minutes per song, or 6 in HQ. The tool is in beta: it works, it is new and it is improving. Tell us what you think at info@obra.art.