What the vocal remover does
A finished song is a single mix: the voice, the drums, the bass and the rest are blended into one track, and an equalizer cannot take the voice out without taking half the music with it. Separating them needs a model that has learned what each one sounds like.
This tool sends your song to our server, where an AI model splits it into two tracks, the vocals and the accompaniment, and gives you both as MP3. It is used for karaoke versions, covers, singing practice, remixes and backing tracks for videos.
How to use it
- Sign in with your free account and drop your song or video on the box above.
- Press Split the song: the file is sent and waits for its turn.
- Wait while the voice and the music are separated, usually two to three times the song's length; the page shows how it is going.
- Listen to the vocals and to the accompaniment on the page.
- Download the track you need, or both.
How it works
The separation is done by Demucs, an open-source model published by Meta AI research, in its hybrid transformer version. It looks at the song both as a waveform and as a spectrogram, the picture of its frequencies over time, and predicts which part of the sound belongs to the voice. Everything that is not the voice becomes the accompaniment.
The song is converted to 44.1 kHz stereo, split on our server CPU one song at a time, and both tracks are encoded as MP3 at 192 kbps. The song is erased as soon as it has been split, and the two parts within 15 minutes.
What to expect
- Very clean: pop, rock, hip-hop and most songs with one clear lead voice.
- Good, with traces: heavy reverb, doubled vocals and backing choirs, which can leave a faint echo of the voice.
- Harder: live recordings, old mono recordings and instruments that sing in the voice range, such as a lead guitar or a violin.
Limits to know
Each account can split 3 songs a day, plus 2 for each friend who joins through your invite link, because each song keeps our server busy for minutes. A song can be up to 5 minutes long and 60 MB, so an MP3 is better than an uncompressed WAV of a long track.
The tool separates sound; it does not grant rights to the music. Practice and private karaoke are fine with any song, but what you publish should use music you own or have permission to use.
Using the tracks in a video
An instrumental makes a good background for a short video, under your own voice. If the voice was recorded on a phone, the background noise remover cleans it first, and the auto caption generator writes the captions once the video is put together.
