How to use Vocal Remover
- Choose a song: an MP3, M4A, WAV, FLAC or OGG file, or a video. Or tap the example to hear it work on a 15-second song.
- Pick what you want: the vocals and the instrumental (a karaoke track), or all four stems: drums, bass, vocals and the rest.
- Tap "Remove the vocals". The first time, the AI (54 MB) downloads once; then the song is separated on your device, 8 seconds at a time, with the time left shown.
- Tap a stem to hear it, switch while it plays, then save each one as a WAV file or all of them in a ZIP.
Why use Vocal Remover
- An AI vocal remover, not a phase trick
- Older vocal removers cancel the centre of the stereo mix and take the drums and bass with it. This one uses Demucs (Hybrid Transformer Demucs by Meta, MIT licence), which listens to the whole song and separates the voice from the music.
- Private: the song stays on your device
- The file is decoded and separated in your browser. Only the AI model downloads, once, from this site; nothing you drop is sent anywhere.
- Vocals, instrumental, or four stems
- Isolate the voice and the karaoke track (vocal isolation and an acapella extractor in one), or extract drums, bass, vocals and everything else on their own, each as a 44.1 kHz stereo WAV.
- Hear every stem before you save
- Each stem has its own waveform; tap one to listen and switch between them while the song plays.
- True progress, and you can stop
- The bar shows the seconds of the song that are done and the time left from the measured speed. Stop early and the part that is finished is kept.
- Any language, one caveat
- The model separates by the sound of a voice and of instruments, not by the words, so it runs on Urdu, Hindi, Arabic or English alike. We measured it on English pop; layered harmony voices and choirs can partly stay in the instrumental.
About this tool
A vocal remover takes a finished song and splits it into the singing and the music behind it. People use one to make a karaoke track for a wedding or a party, to practise a song with the original voice taken out, to hear how the bass line goes, to sample a beat, or to pull the vocals out for a cover or a mashup. For years the only free way was the centre-cut trick, which cancels whatever sits in the middle of the stereo mix: it takes the lead voice down a little, and the kick drum, the bass and half the reverb with it. This tool uses Demucs instead, an open source separation model from Meta research that listens to the whole song and decides, moment by moment and frequency by frequency, what is a voice and what is an instrument.
The whole job happens in your browser. The song is decoded on your device, cut into pieces of about eight seconds, and each piece goes through the model in a background thread while the progress bar counts the seconds done. A recent laptop takes roughly twice the length of the song; a phone takes a few minutes for each minute of music, so on a phone the tool accepts songs up to six minutes and keeps the screen awake while it works. You get the vocals and the instrumental, or four stems, as WAV files that any editor, DAW or karaoke app opens. Separation is never perfect: on a dense mix a little of the voice can stay in the instrumental, and layered harmony voices, choirs and a cappella arrangements are partly treated as music, so some of them can stay in the instrumental too. Use it on music you own or have the right to use; it is not a way to download the stems of someone else's record.
Frequently asked questions
Is this vocal remover free?
Yes. The separation runs on your own device rather than on our servers, so there is nothing to pay per song, no sign-up and no watermark. The first run downloads the AI model (54 MB) once; after that it works from the device.
Is my song uploaded?
No. Your browser decodes the file and the AI separates it on your device. Only the model is downloaded, from this site, the first time. We never see the song.
How good is it? Will some of the voice stay in the instrumental?
Demucs is one of the best open separation models (its authors report 8.8 dB of signal-to-distortion on the standard test set). On the English pop songs we tested, the instrumental was clean enough to sing over in 9 cases out of 10. On a very dense mix, or where the singer is doubled and drowned in reverb, a faint trace of the voice can stay, and layered harmony voices or a choir are partly treated as music and can stay in the instrumental.
How long does it take?
About twice the length of the song on a recent laptop, counted in the progress bar as it goes. Phones take a few minutes per minute of music, so on a phone the tool takes songs up to six minutes and keeps the screen awake. Keep the page open; you can stop early and keep the part that is done.
What is a stem?
One part of a song on its own: the drums, the bass, the vocals, or everything else (guitars, keys, strings, synths). The "4 stems" option gives you all four; "Vocals + instrumental" gives the voice and the three others mixed together. People also call the voice on its own an acapella, and the tool that pulls it out an acapella extractor or a vocal isolation tool.
Can I make a karaoke track for a wedding or a party?
Yes. Choose "Vocals + instrumental", separate the song, and save the instrumental as a WAV. It plays in any music app, and the vocals file lets you check the melody first.
Does it work on Urdu, Hindi, Arabic or Punjabi songs?
The model separates by the sound of a human voice and of instruments, not by the language, so it runs on Bollywood, Coke Studio and Arabic pop the same way. We have measured its quality on English pop so far; on an a cappella nasheed with layered voices, part of the harmonies went into the instrumental. Instruments it has heard less often, such as the harmonium or the sitar, mostly go into the "other" stem.
Which files can I use, and what do I get back?
MP3, M4A, AAC, WAV, FLAC, OGG, Opus, and the sound of MP4, MOV or WebM videos. You get 44.1 kHz stereo WAV files, one per stem, or a ZIP of all of them.
Can I use this on any song I like?
Use it on music you own or have permission to use: your own recordings, royalty-free music, or songs you have a licence for. Removing the vocals from a commercial record for your own practice is one thing; sharing or selling the result is up to the copyright owner.
Gaane se awaaz kaise hataye? (How do I remove the voice from a song on my phone?)
Gaana chunein, "Remove the vocals" dabayein, aur AI phone par hi music aur awaaz alag kar deta hai; phir instrumental save kar lein. Choose the song, tap "Remove the vocals", and the AI separates the music and the voice on the phone itself; then save the instrumental.
Built and maintained by Saad Naeem
AI engineer, Doha · Updated