Vocal remover: split vocals and instrumental

Karaoke tracks and a cappellas in a minute

Choose a song and press Separate. An AI model trained to tell voices from instruments listens to the music on your device and splits it into two tracks:

• Instrumental: the song without the singer, for karaoke, covers or practising an instrument.
• Vocals: the voice alone (a cappella), for remixes, transcribing lyrics or learning a melody.
• Both: download the two tracks at once.

It works with MP3, WAV, M4A, FLAC, Ogg and the audio of videos (MP4, MOV, MKV, WebM), up to 10 minutes. The first time, the page downloads the model (about 39 MB); after that it is cached and the separation starts right away.

Tips for a good result

• Studio recordings separate best; live recordings with crowd noise and lots of reverb leave traces.
• Backing vocals and choirs count as vocals; a saxophone or a lead synth can sound like a voice to the model.
• Choose WAV if you will edit the tracks further, MP3 to share them.

The song is turned into a spectrogram, a picture of how loud each frequency is over time. Spleeter, an open-source model from Deezer (MIT licence), estimates which parts of that picture belong to the voice and which to the instruments. Those estimates become a mask applied to the song, and the vocals are turned back into sound; the instrumental is the song minus the vocals.