Vocal Remover: Remove Vocals From a Song

How AI separates the vocals from the music

A song is a mix: the vocals and the instruments are added together in a single file, and there is no way to simply "peel" them apart. AI works differently. It slices the sound into thousands of tiny pieces of time and frequency, then estimates, for each one, how much belongs to the voice and how much belongs to the instruments. The model was trained on thousands of songs whose vocals and backing tracks were available separately, so it has learned to recognize the sound of a singing voice, its attacks and its vibrato, even when it is buried under guitars or drums. You get two tracks, the instrumental and the isolated vocals, which add back up to exactly the original song. It is the same principle as professional stem separation software, except that it runs here in your browser, without uploading the file.

Make a karaoke track, a backing track or an instrumental

The instrumental you get is a ready-made backing track: sing karaoke at home or at a party, rehearse for a talent show, a school play, a church service or an audition, or practice singing over the real music. Download it as an MP3 to play it anywhere (phone, Bluetooth speaker, TV through a USB stick), or as a WAV to edit it in audio or video software without any further quality loss. If the song is too high or too low for your voice, run the instrumental through the "Pitch Changer" to shift it by a semitone or two. To keep only a verse or the chorus, trim it with the "MP3 Cutter". The vocals-only track, for its part, helps you catch lyrics that are hard to understand, study how a singer performs a song or prepare a remix.

AI mode or fast mode: which one should you choose?

Fast mode uses an old trick from audio software: the lead vocal is almost always placed in the center of the mix, identical in both speakers, while the instruments are often spread to the left and right. By removing what both channels have in common between 250 Hz and 7 kHz, much of the voice disappears instantly, while the bass and kick drum stay. But everything in the center goes too (snare, solo, piano), the reverb on the voice remains audible, and a mono recording cannot be processed this way. AI mode does not depend on placement: it recognizes the voice itself, keeps the stereo image of the instruments and gives you usable isolated vocals. It only requires downloading the model once (about 40 MB), then a few seconds to a minute or two of processing depending on your device.

The limits of vocal separation, and copyright

No separation is perfect. Depending on the song, a faint echo of the voice may remain in the instrumental, especially reverb and backing vocals, and the isolated vocals may keep a little drums or sound slightly "underwater". Songs with lots of backing vocals, heavily processed voices (effects, strong Auto-Tune) or instruments close to a voice, such as a saxophone or a violin, separate less cleanly. Start from the best-quality file you have: a heavily compressed MP3 gives a less clean result. Finally, removing the vocals does not change the rights: the instrumental of a commercial song is still protected by copyright. It is fine for personal use, such as singing at home or practicing, but posting it on YouTube or TikTok, or playing it in public, requires permission from the rights holders.