Make a karaoke or instrumental version of a song by removing the centre-panned lead vocal, or pull out an approximate acapella. Compare the original, instrumental and vocal tracks, then download them as WAV or MP3. Your music is processed on your device and never uploaded.
Method
Choose a song or try the demo.
0:00 / 0:00
Original
Instrumental
Vocals
Advertisement
Advertisement
How to use the Vocal Remover
Choose a song: drop a stereo MP3, WAV, M4A, OGG or FLAC file on the box, or press Try a demo song.
Wait a few seconds while the vocals are separated. The progress bar shows how far it is.
Listen: press play and switch between Instrumental, Vocals and Original, or click a waveform to jump to that point.
Fine-tune: lower Keep bass below if the bass sounds thin, widen Centre focus if the voice is still audible, or try Classic mode.
Download the instrumental or the vocals as WAV or MP3.
Centre-channel removal explained
A stereo song has a left (L) and right (R) channel. A sound panned exactly to the centre, like most lead vocals, is equally loud and in phase in both. Mid/side processing splits the mix into two parts:
Mid (centre) M = (L + R) ÷ 2: everything common to both channels.
Side S = (L − R) ÷ 2: everything that differs between the channels.
The classic karaoke trick outputs only the side signal, L − R, so anything identical in both channels cancels out. The problem is that bass, kick drum and snare are centred as well, so this tool subtracts only the part of the mid signal above the Keep bass below frequency: left′ = L − HP(M) and right′ = R − HP(M), where HP is a zero-phase high-pass filter. Everything below the cutoff, and every side-panned instrument, is kept. The removed part, HP(M), is the vocal track.
Smart mode goes one step further. It splits the song into short overlapping windows (about 46 ms) and about 1,000 frequency bands, and compares left and right in each band using the similarity 2·Re(L·R̄) ÷ (|L|² + |R|²), which is 1 only when both channels are identical. Only bands that are nearly identical are treated as vocal, so a guitar panned partly to the left keeps its centre component instead of being thinned out. Centre focus sets how close to identical a band must be.
Getting the best results
Problem
What to try
Voice still clearly audible
Widen Centre focus, or switch to Classic mode. Reverb on the voice cannot be removed.
Bass or kick too weak
Raise Keep bass below to 150 to 200 Hz.
Cymbals or hi-hats sound dull
Set Keep highs above to 10 or 12 kHz.
Instruments sound thin or watery
Narrow Centre focus or lower the removal strength to 70 to 85%.
Nothing changes at all
The file is probably mono or has the voice panned to one side.
Use the highest-quality source you have. Low-bitrate MP3s (128 kbps and below) often use joint-stereo tricks that blur the difference between channels, which leaves more vocal behind than a WAV, FLAC or 320 kbps file.
Frequently asked questions
How does the vocal remover work?
In most studio mixes the lead vocal is panned to the centre, so it is identical in the left and right channels, while many instruments are panned to the sides. Subtracting the centre part removes the vocal. Classic mode does this with L − R phase cancellation; Smart mode analyses thousands of frequency bands per second and removes only the bands that are the same in both channels.
Why can I still hear some of the vocals?
Reverb, echo and stereo effects on the voice are often spread across both channels, so they are not centred and cannot be cancelled. Backing vocals panned to the sides also stay. Try widening Centre focus or switching to Classic mode, but expect a faint vocal 'ghost' on many songs.
Why did the bass and drums get quieter?
Bass, kick drum and snare are usually centred too, so they are removed along with the voice. Keep bass below protects everything under the chosen frequency, 120 Hz by default. Raise it to 150 to 200 Hz to keep more of the bass, at the cost of some low voice.
Does it work with mono songs?
No. In a mono file the voice and the instruments are mixed into one signal, so there is no difference between channels to work with. Old recordings from before the late 1960s are often mono or have the voice panned to one side.
How good is the acapella (vocals) track?
It is an approximation: it contains everything that is centred, so the vocal comes with some of the snare, bass or other centred instruments. It works best for songs with a sparse arrangement. AI stem separation models give cleaner results but need much more computing power or a server upload.
Is my music uploaded?
No. Decoding, separation and encoding all run in your browser with the Web Audio API and JavaScript, so your files stay on your device.
Can I use the karaoke tracks publicly?
Removing vocals does not change copyright. Instrumental versions of commercial songs are fine for personal practice and private karaoke; public performance or publishing usually needs permission or a licence.