How to remove vocals from a song for karaoke

A vocal remover can turn a stereo song into a sing-along track in seconds. It is not magic, and knowing the mechanism explains the odd artefacts.

The basic idea: centre-channel cancellation

In most stereo mixes, the lead vocal is placed dead centre — the same signal in both the left and right channels. Instruments are spread around the stereo field: guitars to one side, keyboards to the other. If you subtract one channel from the other, anything identical in both cancels out, while anything panned left or right survives.

That is the whole trick behind a classic karaoke track. It is simple, fast, and needs no artificial intelligence or downloaded model. The vocal remover on this site uses exactly this method, running locally in your browser.

Why the bass and drums fade too

Here is the catch. Producers usually put the bass line, kick drum and snare in the centre as well, because a solid centre anchors a mix. Those sounds are identical in both channels, so the cancellation removes them along with the voice. The result often has thin low end and a weak rhythm section.

This is not a bug in the tool — it is a direct consequence of how the method works. Anything centered will disappear; anything panned will remain. If a particular song still has audible vocals afterwards, it is probably because the voice was doubled, panned or treated with stereo effects.

It needs a real stereo file

Cancellation compares left against right, so a mono file has nothing to subtract. A mono recording — including every AMR voice note — would come out silent, which is why the vocal remover rejects those files rather than producing a useless result.

You need a genuine stereo song: MP3, WAV, M4A or FLAC with two distinct channels. Check in your player or editor that the file is stereo before you start. A "fake stereo" file, where one channel was simply copied from mono, will also cancel to silence.

How this differs from AI separation

AI stem separation uses a trained model to identify and isolate vocals, drums, bass and other instruments as separate tracks. It can preserve the bass and drums while removing the voice, which makes it much more precise. The cost is a large model download, more processing time, and a dependency on somebody else's trained network.

Centre-channel cancellation is the opposite: small, instant, private and predictable. It will never be as clean as AI separation, but it needs no model, no upload and no account. For a quick karaoke track from a song you already have, that trade is often perfectly reasonable.

Why AMR will not work here

It is worth stating plainly because this tool sits alongside AMR converters on this site: an AMR voice note cannot be turned into a karaoke track. AMR is mono by design — one channel, no left or right difference — so subtraction has nothing to operate on. The result would be silence, not an instrumental.

If you uploaded an AMR file expecting a vocal remover, what you actually need depends on the goal. To make a song's vocals quieter you need the original stereo song, not the voice recording. To make a voice recording itself easier to work with, try the noise reducer or the volume normalizer instead, or use the transcription tool to get the words as text.

Getting the best result

Choose songs with a prominent centred vocal and plenty of stereo instrumentation — the separation will be cleaner. Avoid mono masters and old recordings with everything panned to the middle. Export at a decent bitrate; the tool can write MP3, WAV, M4A or FLAC, and WAV is a good choice if you plan to add a bass line back in an editor.

One last point that matters more than the technique: only process music you have the right to use. Making a personal sing-along track is one thing; distributing a song with the vocals removed is a copyright issue regardless of how the track was made.

Related reading

Put it into practice

Every tool on this site runs locally in your browser. Convert AMR to MP3, transcribe to text, or see all tools.