AI Vocal Remover: split the voice from the music
Upload any song and get two files back — the vocal on its own and the instrumental on its own. Not a filter that scoops out the middle of the mix, but a real separation.
What it does
Two tracks out, not one
The old way to remove vocals was a phase trick: flip one stereo channel against the other and anything sitting dead centre cancels out. It does silence the singer, and it also guts the kick, the snare and the bass, because those live in the centre too. What is left sounds hollow.
This does something different. The model identifies each part of the recording and rebuilds it as its own audio file, so removing the voice does not cost you the low end. You get both halves — the instrumental to sing over, and the isolated vocal, which is the half people usually want for remixing and sampling.
How clean it comes out depends on the recording. A modern, well-separated mix splits almost invisibly. Heavy reverb, a vocal buried under a wall of guitars, or a mono live capture are the hard cases, and you may hear traces of one part in the other.
How it works
How to remove vocals from a song
About ninety seconds from upload to download.
- 01
Upload the track
MP3, WAV, FLAC or M4A up to 20MB. A studio recording, a live capture, a phone voice memo — anything with a voice over music.
- 02
The parts are separated
The vocal and the accompaniment are identified and reconstructed as two independent files. Nothing is filtered out of the original; both are built fresh.
- 03
Play and download both
Preview each one in the browser, then download whichever you need. They are the same length and start at the same point, so they line up in a DAW.
Hear it
The same thirty seconds, before and after
The full mix, then the two files it separates into. All three waveforms are drawn to the same scale, so a quieter part really is quieter.
Original mix
98–128s
Vocal, on its own
separated
Instrumental, on its own
separated
What you get
What makes a good AI vocal remover
Vocal isolation and instrumental in one run
Both halves come back from a single upload. Most tools that advertise vocal removal hand you the backing track and throw the voice away — the isolated vocal is usually the more useful of the two.
The low end survives
Phase cancellation takes the kick and bass out along with the singer, because centre-panned content all cancels together. Separation rebuilds each part instead, so the instrumental still has a bottom.
- invert one channel
- centre content cancels
- bass and drums go with it
- each part identified
- rebuilt as its own file
- the mix stays intact
An online vocal remover with nothing to install
Runs in the browser on any machine — no download, no plugin, no GPU. MP3, WAV, FLAC, M4A, AAC and OGG all work, up to 20MB per file.
Or take the song apart instrument by instrument
When two tracks are not enough — you want the drums without the bass, or just the guitar — the stem splitter separates by instrument, and targeting a single stem costs well under a full split.
Two tracks not enough? The stem splitter pulls each instrument out separately, and you can ask for just one of them for a fraction of a full split.
Who uses it
What people do with the two halves
Karaoke and backing tracks
Drop the lead vocal and keep the band. The obvious use, and the one the low end matters most for.
Acapellas for remixing
The isolated vocal is the starting point for every remix and mashup — the part almost nobody is given.
Practising along
Mute the part you are learning and play it yourself over the rest of the recording.
Sampling a phrase
Lift a vocal line out from under the arrangement instead of filtering and hoping the bleed is not audible.
Content and voiceover beds
Turn a track with singing into an instrumental you can talk over without the two voices fighting.
Checking a mix
Hearing the vocal alone tells you things about a production that the full mix hides.
FAQ
Vocal remover questions
UVR is a free open-source desktop app. It is excellent, and if you have a decent machine, are willing to install Python-based software and pick models yourself, it will do everything this does and more — for nothing. The trade is setup and hardware: UVR runs locally and wants a capable GPU for reasonable speed. This runs in a browser on any machine with no install, and the separation happens on our side. If you split tracks every day, learn UVR. If you need two files this afternoon, use this.
Not unlimited, but a new account starts with enough credits for two separations, so you can run it on a real track before paying anything. After that it is priced per run rather than by subscription — there is no plan to cancel if you only needed it once.
Good enough to use on most modern, well-recorded material. It is a reconstruction rather than the original multitrack, so expect some bleed on dense mixes, heavy reverb and anything where the voice and an instrument occupy the same frequency range. Sparse arrangements come out cleanest; wall-of-sound productions are the hardest case.
Yes, both, from the same run. They are the same length and aligned to the original, so dropping them onto two tracks in a DAW lines them up exactly.
Usually reverb. The dry voice and its reverb tail are different things, and a long tail smeared across the mix is genuinely ambiguous — some of it reads as part of the room rather than part of the singer. Backing vocals doubled with instruments cause the same effect.
MP3, WAV, FLAC, M4A, AAC and OGG, up to 20MB per file. Lossless sources separate marginally more cleanly than a low-bitrate MP3, but the recording itself matters far more than the format.
Around ninety seconds for a typical song. You can leave the page while it runs — finished jobs also appear under History.
That is a different job: this splits a song two ways, vocal and everything else. The stem splitter separates by instrument — drums, bass, guitar, piano, strings and more as individual files — and if you only want one of them, targeting a single stem costs considerably less than a full split.
You can upload it, and there is nothing wrong with separating a record to practise along with it, study it or work on a remix privately. Releasing or monetising anything built from the result is a different matter and needs permission from the rights holder — separating a track does not create a new copyright for you.
It will run, but expect a rougher result. Crowd noise, room reflections and a mono source all make the parts harder to tell apart. A clean studio recording is the best case by a wide margin.
Take the voice out of a track
Upload a song and get the vocal and the instrumental as separate files.