Vocal Remover
Vocal Remover cancels whatever sits in the centre of the stereo image, which on most recordings is the lead vocal. It is a fast way to make a rough backing track. It is not AI separation, and this page is explicit about what it can and cannot do.
Your track stays on your device.
The processing runs in this tab. That is also why it is phase cancellation rather than a trained separation model — the good models need a server with a GPU, and this needs two channels and some arithmetic.
Learn how files are processed →- Requires
- A stereo file — mono cannot work
- Method
- Centre-channel phase cancellation
- Removes
- Whatever is centred, vocal or not
- Output
- Same format as the source
How to use Vocal Remover
- 01Upload
Load a stereo track
The file must be stereo — the technique works on the difference between the two channels.
- 02Customize
Set the strength
Full cancellation removes the most vocal and the most backing. Less leaves more of both.
- 03Process
Export
Download the result and listen on headphones before relying on it.
What is Vocal Remover?
Vocal removal by phase cancellation is an old studio trick, and it works because of how records are mixed. The lead vocal is almost always placed in the exact centre of the stereo image, which technically means the identical signal is sent to both speakers.
Anything identical in both channels vanishes when one channel is subtracted from the other. That single operation is the whole technique, and it takes milliseconds.
What it leaves behind is everything that differed between the channels — the guitars panned left, the keys panned right, the reverb, the room. That is why the result sounds wide and hollow: you are listening to the sides of the mix with the middle removed.
Supported formats
- MP3
- WAV
- M4A
- OGG
- FLAC
- Maximum file size —
- bounded by your device's memory
- Processed by —
- FFmpeg (WebAssembly), on your device
Limitations
Worth knowing before you start, rather than after.
- This is phase cancellation, not AI separation. It removes whatever is centred in the stereo image — usually the vocal, but often the kick, snare and bass too.
- It cannot work on a mono file. There is no difference between the channels to subtract.
- Vocals with heavy reverb, or vocals that are not centred, survive the process largely intact.
Why a free browser tool cannot match a trained model
Phase cancellation does not know what a voice is. It knows what is centred. Every decision it makes is geometric, and the consequences follow directly: centred instruments go with the vocal, off-centre vocals stay, and no amount of tuning changes that because there is no analysis happening at all.
Trained separation models work differently. They have learned what a human voice looks like as a spectrogram and reconstruct each part separately, which is why they can pull a centred vocal out while leaving the kick drum intact. They also need a GPU and several seconds to minutes per track, which means a server, which means a running cost per use.
That is the honest trade this tool represents. It is instant, private and free, because it is arithmetic on two channels. It is worse than the paid alternatives, because it is arithmetic on two channels.
For practising a part, checking an arrangement, or making a rough karaoke track for a party, it is usually enough. For anything released or performed, it is not, and no setting on this page will make it so.
One practical tip: if the result is close but the vocal ghost is distracting, running the output through the Equalizer and cutting around 1–3 kHz often pushes the remainder far enough back to be usable.
Your file, and what the page does
Your file is never uploaded. Vocal Remover reads it into this browser tab and processes it with FFmpeg compiled to WebAssembly on your own device. There is no upload endpoint in the application and no copy on any server — closing the tab discards it.
That is also why processing takes longer here than on a site that uploads: your device is doing the work rather than a rack of servers. The trade is that the file never leaves it.
What the page does send or fetch
- The processing engine, once
- The first time you use any video or audio tool, the browser downloads the FFmpeg WebAssembly build (~30MB) from the jsDelivr CDN and reuses the cached copy afterwards. That is a file coming to you — no part of your own file is in the request.
- Ads on the page
- The site is funded by advertising, so the page loads Google AdSense. That carries the usual web basics — IP address, browser, referring page — as on any ad-supported site.
- An anonymous usage counter
- When Vocal Remover finishes we record that the tool ran, whether it succeeded and how long it took. No file data, no identifier, no cookie.
Your file is not on that list. How this works.
Frequently asked questions
In most stereo mixes the lead vocal is panned dead centre, meaning it is identical in the left and right channels. Subtracting one channel from the other cancels anything identical in both — the vocal disappears, and so does everything else that was centred.
Because they are usually centred as well. Kick, snare and bass sit in the middle of almost every mix, so they cancel along with the vocal. This is the fundamental limitation of the technique and no setting avoids it entirely — lowering the strength keeps more of them, at the cost of keeping more vocal.
Three common reasons: the vocal is not perfectly centred, it has stereo reverb or double-tracking that is different in each channel, or the file is mono to begin with. Reverb in particular survives, so you often get a track with no dry vocal but an audible ghost of it.
No. Mono means both channels are identical, so subtracting one from the other leaves silence. The tool checks for this and says so rather than handing back an empty file.
No. Trained separation models analyse the sound and reconstruct the parts individually, and they get much better results — they also need a server with a GPU. This runs in your browser in seconds and costs nothing, with the trade-offs described above.
Often yes for practice, rarely for performance. Test it on headphones; the artefacts are much more obvious there than on speakers.
Yes — Vocal Remover is completely free. There's no account, sign-up, watermark or limit.
No. It runs in any modern browser — Chrome, Edge, Safari, Firefox, Brave — with nothing to download or install.
Need a backing track to practise over?
Load a stereo song and hear what centre cancellation gives you.
Edit any media with a right-click
Add Dexta Studio to Chrome and open any image, video, audio or PDF straight into the right tool — free, nothing uploaded.