Free · Unlimited · Nothing uploaded

Vocal remover & instrumental maker

Turn a stereo song into an instrumental or karaoke track — adjustable strength, bass preservation, instant preview. No upload, no daily cap, no account.

🎤

Drop a song here or click to browse

MP3 · WAV · M4A · OGG — a stereo file works best, processed on your device

🔒 Decoding and processing run entirely in your browser — your audio never leaves your device.

How it works

🎚️

Frequency by frequency

A short-time Fourier transform splits the track into narrow bands. In each one it measures how much the two channels agree — a centred vocal appears identically in both — and removes only the shared part, in proportion to how centred it is.

🎧

It stays in stereo

Bands that are not centred are never touched, so the guitars keep their positions. That is the whole difference from the old left-minus-right trick, which flattens the entire mix to mono to get at the vocal.

🎤

Both stems

The vocal is exactly what was subtracted, so you get it on its own for free — instrumental plus vocals adds back up to the original, sample for sample.

🔊

The bass survives

Kick and bass sit dead centre too. Everything below the line you set is left completely alone, which is the difference between an instrumental and a telephone.

▶️

A player, not two buttons

Click the waveform to jump to the chorus, then switch stems without losing your place. You cannot judge a separation from the intro.

⚡

Seconds, not minutes

The heavy pass runs in a worker with the two channels packed into one transform, so a four-minute track separates in about a second and a half without freezing the tab.

Unlike most free vocal removers, which cap you at a couple of songs a day or watermark the preview until you pay, this runs the whole pipeline — decode, analyse, separate, encode — locally in your browser. Nothing is uploaded and nothing is counted. No file size games, no daily counter. Once you've got your track, trim it with the Audio Cutter, even out the levels with the Audio Normalizer, or change its speed with the Speed & Pitch Changer.

Why the old trick works and mostly does not

The classic vocal removal technique is centre-channel cancellation: subtract the right channel from the left. Because lead vocals are usually mixed at equal level in both channels, they cancel out, while anything panned to one side survives.

The results are usually disappointing, for several reasons. Everything else placed centrally goes too — kick drum, bass, snare and often the lead instrument, since centre placement is standard for the rhythm section. The output is also mono by construction, having collapsed two channels into one.

Worse, vocals are rarely purely central. Reverb and delay on a voice are typically stereo, so the dry signal cancels while its reverb tail remains, leaving a ghostly wash where the vocal was. Double-tracked or harmonised vocals panned apart do not cancel at all.

What source separation does instead

Modern separation uses neural networks trained on large collections of music where the individual stems were available, learning the spectral and temporal signatures that distinguish a voice from a guitar. Rather than exploiting stereo placement, they identify sources by what they sound like, which is why they work on mono material and on centrally placed instruments.

The practical quality is far higher, and separated stems are usable for rehearsal, transcription, DJ work and study. It is not perfect: artefacts appear where sources overlap heavily in frequency, and the residue is often described as watery or smeared. Dense mixes, heavy compression and unusual production styles all reduce accuracy.

Source material quality matters more than people expect. Separation from a low-bitrate MP3 works on data that already had high-frequency content discarded, and the model reconstructs from a degraded input. Starting from a lossless file produces noticeably better stems.

Copyright and legitimate uses

Separating stems does not create a new work you own. The recording remains copyrighted, and both the composition and the specific recording carry separate rights. Producing an instrumental version of a commercial track is making a derivative work, which requires permission.

Uses that are generally fine: private practice and rehearsal, studying an arrangement, transcription, and working with material you own or that is properly licensed. Karaoke for private use sits comfortably here in most interpretations.

Uses that are not: publishing an instrumental version, distributing stems, using a separated backing track in a monetised video, or releasing a remix without a licence. Platform content-matching systems detect separated stems reliably, since the underlying recording is still present, so a video using one will typically be claimed or muted. Commercial use requires clearing rights with the publisher and the label separately.

Vocal remover FAQ

How does vocal removal work?

Most studio recordings mix the lead vocal dead-center, panned identically into the left and right channels. This tool subtracts the right channel from the left — anything identical in both channels, usually the vocal, cancels out. Bass preservation mixes some low end back in since kick/bass are often centered too.

How does it separate the vocal?

The default Studio method runs a short-time Fourier transform and works frequency by frequency. For each narrow band of each moment it measures how much the left and right channels have in common — a centred source appears identically in both, a wide one does not — and removes only that shared part, in proportion to how centred it is. Bands that are not centred are left completely alone, which is why the guitars keep their positions and the result is still in stereo. The vocal stem is simply what was taken out.

Why isn't the vocal removal perfect?

Because this is signal analysis, not a trained model. It works on where a sound sits in the stereo field, so it separates cleanly when the vocal is centred and the music is not — which describes most studio pop — and struggles with a mono-ish mix, a live recording, or a vocal that has been widened. What it will not do is ration you: no daily cap, no upload, no account, no watermark.

Can I get the vocal on its own?

Yes — Vocals only in the player, and it exports too. It is exactly what was subtracted, so instrumental plus vocals adds back up to the original.

What do Focus and Protect bass do?

Focus is how certain the tool has to be that a sound is centred before it touches it. Turn it up to keep more of the music (and leave more vocal); turn it down to strip harder, at the risk of dulling instruments that sit near the middle. Protect bass leaves everything below the chosen frequency completely alone — kick and bass are centred too, and removing them is what makes a naive karaoke track sound like a telephone.

Does it work on any song?

It needs a stereo file. The whole method is about what the two channels have in common, and a mono file has only one channel, so there is nothing to compare. Most commercially released music qualifies.

Can I hear the same part of each version?

Yes, and that is the point of the player. Click anywhere on the waveform to jump there, then press 1, 2 and 3 — or the tabs — to switch between original, instrumental and vocals without losing your position. Judging a separation from the intro tells you nothing; you need the chorus.

Is my audio uploaded anywhere?

No. Decoding, processing and exporting all happen locally using the Web Audio API — nothing leaves your device, no daily limit, no account.

Why does the classic left-minus-right trick sound so bad?

Because it is indiscriminate. Subtracting one channel from the other removes everything panned centrally — kick, bass and snare along with the vocal — and what comes out is the same signal in both speakers, so the mix collapses to mono. The vocal's stereo reverb is not centred, so it survives as a ghostly wash. It is still available here as Fast, because on a dry, hard-centred mix it is fine and it is instant, but it is no longer the default.

Can I release a track with the vocals removed?

Not without permission. An instrumental version is a derivative work, and both the composition and the recording carry rights. Private practice and transcription are generally fine; publishing or monetising is not.