Remove vocals and make karaoke online

Frequently asked questions

Is my song uploaded anywhere?

No. The audio is decoded and processed in your browser, in a background worker, and never leaves your device. There is no account. You only get a file when you press download.

Is this AI stem separation?

No, and it is not built. This is classic centre-channel processing: it finds what is identical in the left and right channels and takes it out (or keeps only it). It cannot truly separate vocals, drums, bass and other instruments the way an AI model such as Demucs can. Anything else mixed in the centre inside the voice band, such as a snare or a lead guitar, goes with the voice.

How is this better than simply subtracting one channel from the other?

The plain subtraction removes everything in the centre, including bass and kick, and gives a mono file. Here the centre is taken out only between the two band edges you choose (120 Hz to 8 kHz by default) with steep filters, so centred bass and the highest frequencies stay, and the result is stereo.

What do the sliders do?

Band starts at and Band ends at set the voice range. Strength is how completely the band is removed (or, when keeping the centre, how far the rest is turned down). Centre width, for keeping the centre, lets some stereo-spread sound through: 0% keeps only what is identical in both channels. At 0% width and 100% strength, the instrumental and the vocal mix add back up to the original.

Why does it need a stereo file?

The method uses the difference between left and right. A mono file, or a stereo file whose two channels are identical, has nothing to work with, and the page tells you when it sees one. A file that is almost mono gets a warning too.

Does it change the sample rate or clip loud songs?

The result keeps the sample rate of the file you loaded. A limiter, not hard clipping, keeps the result under -1 dB full scale, and the page says how much it had to turn peaks down. Comparing Original and Result switches instantly at the same position, with loudness matched while you listen; the saved file itself is not level-adjusted.

Which formats can I save?

WAV 16 or 24-bit, FLAC 16 or 24-bit (lossless, written by a small encoder built into this page), and MP3 at a constant bitrate. You can also load a video (MP4, MOV, WebM and others); the sound is taken from it and you save audio only.

Which songs work best?

Tracks with the vocal dead-centre and instruments spread across the stereo field. Heavy reverb on the voice, a vocal doubled to the sides, or a very narrow mix leave more behind. Expect a rough karaoke track, not a studio acapella.