Vocal remover. Karaoke tracks and acapellas.
Removes the lead vocal from a stereo song, or isolates it, entirely in your browser with nothing uploaded. It works by cancelling whatever is panned dead centre, and keeps the bass intact so the backing track does not come out thin. Free, no signup.
drop a stereo song
MP3, WAV, M4A, FLAC — must be stereo, not mono.
At a glance
| Method | Centre-channel cancellation (mid/side), not AI stem separation |
|---|---|
| Input | Stereo audio — MP3, WAV, M4A, FLAC, OGG. Mono files will not work |
| Outputs | Instrumental (vocals removed) or acapella (centre isolated) |
| Bass handling | Adjustable crossover keeps low end in the instrumental |
| Output format | MP3 (128–320 kbps) or WAV |
| Practical size limit | ~500 MB per file (browser memory) |
| Uploads your file | No — runs entirely in the page |
| Signup required | No |
How it actually works
In almost every commercial mix the lead vocal is panned dead centre, which means it appears identically in the left and right channels. Subtract one channel from the other and anything identical in both cancels out — the vocal disappears, and what is left is whatever was panned to the sides: guitars, keys, reverb, backing vocals.
That is the entire trick, and it has been how karaoke machines worked since long before anyone said "AI". It is fast, it runs on any device, and on a well-made stereo mix it is genuinely effective.
The catch is the low end. Kick drum and bass are also usually centred, so a naive subtraction strips them out too and leaves a thin, hollow backing track. This tool splits the signal at a crossover — 200 Hz by default — and puts the bass back below that point while cancelling only above it. That one change is the difference between a result you would actually sing over and one that sounds broken.
When it works well, and when it doesn't
It works well on: most pop, rock, country and hip-hop mixes from the last forty years; anything with a clearly centred lead vocal and instruments spread across the stereo field; live recordings with a centred PA feed.
It works badly on:
- Mono files. There is no difference between the channels to exploit, so there is nothing to cancel. The tool will tell you rather than hand you silence.
- Vocals with heavy stereo reverb or double-tracking. The dry vocal is centred but the reverb tail is not, so you get a ghostly wash left behind.
- Anything else panned centre. Snare, bass and often the lead guitar sit in the middle too, so they get thinned along with the vocal.
- Mixes that were mastered narrow, or old recordings that are nearly mono — the less stereo information there is, the less there is to work with.
Being honest about this matters more than overselling it. If you need broadcast-quality stem separation, you need a machine-learning model like Demucs, which requires a large download and a lot of processing. This is the fast, private, good-enough version that runs in a tab.
Settings that matter here
Keep bass below is the setting that most affects how usable the instrumental sounds. 200 Hz is the sweet spot for most music: it preserves kick and bass while still cancelling the vocal fundamental. Push it to 300 Hz if the track sounds thin, drop it to 120 Hz if you can still hear too much of the vocal's body. Set it to zero for pure cancellation, which is cleaner on the vocal but hollow underneath.
Boost the result is on by default because cancellation typically drops the level by 6 dB or more, and an unboosted result sounds broken when it is merely quiet. If you are going to process it further, turn it off and normalize deliberately at the end instead.
For the acapella mode, expect a rougher result than the instrumental. Isolating the centre keeps the vocal but also keeps centred drums, so you tend to get vocals with a ghost of the kick and snare behind them. It is useful for reference and practice, less so for building a new track.
What you lose
Stereo width, mostly. Cancellation collapses a lot of the spatial information, so the instrumental is narrower than the original mix even though it is output as stereo.
Some of the instruments, some of the time. Anything sharing the centre with the vocal gets attenuated with it, so a centred lead guitar or a prominent snare will sound quieter or slightly phasey.
Fidelity, if you output to MP3. You are re-encoding audio that was already lossy, so use 320 kbps if the result is going anywhere important, or WAV if it is going into a DAW for further work. Trimming to the section you need before processing also saves time on long files.
FAQ
Does this actually remove vocals, or just make them quieter?
On a well-mixed stereo track it removes them almost entirely. On tracks with heavy reverb or double-tracked vocals it reduces them substantially but leaves a residue, because the parts of the vocal that are not dead centre cannot be cancelled.
Why does it say my file is mono?
Centre cancellation works by subtracting the right channel from the left. In a mono file both channels are identical, so subtracting them gives silence — there is no stereo information to separate. You need a genuine stereo mix.
Is this the same as AI stem separation?
No. Tools like Demucs and Spleeter use trained models to identify and separate instruments, and get better results on difficult material. They also need a large model download and much more processing. This is centre cancellation: instant, private, and effective on most conventional mixes.
Why does the instrumental sound thin?
Because kick and bass are usually centred too, so a plain cancellation removes them. Raise the 'keep bass below' crossover to 300 Hz and the low end comes back. That setting exists specifically to fix this.
Can I get a clean acapella out of it?
Partly. Isolating the centre keeps the vocal but also keeps anything else centred, typically kick and snare. It is good enough for learning a part or checking lyrics, not usually clean enough to build a new production around.