Stem Separation
Remove Vocals from Any Song
Need a clean instrumental for karaoke, a remix, or a backing track? KROMA separates your song into six stems — mute the vocal track and the band stays untouched. Vocal extraction runs a Roformer-family model, the architecture leading public separation leaderboards.
How it works
- 1
Drop in your song
WAV or MP3 — studio masters, live recordings, even AI-generated tracks all work.
- 2
We split it into stems
Six independent tracks: vocals, drums, bass, guitar, piano, and everything else — each with its own fader.
- 3
Mute the vocals
Pull the vocal fader to zero for a pure instrumental — or drop it to −12 dB for a soft 'guide vocal' feel.
- 4
Download the instrumental
Get the mix back as a clean master — karaoke-ready, remix-ready, no vocal bleed on the band.
Sounds KROMA can isolate
A real fader, not a phase-cancellation trick
Old-school vocal removers subtract the centre channel — which also guts any instrument panned to the middle, like the bass and the kick. KROMA uses source separation instead: a model trained on millions of mixed tracks reconstructs the band underneath the vocal, so the instrumental keeps its weight.
Because the vocal is its own stem, you get options a simple remover cannot offer: mute it completely, keep it low as a guide, or remove only a backing-vocal layer by ear.
More than karaoke
The same separation runs both ways. Flip it and isolate the vocal only for a clean acapella — or mute the drums and bass to practice over your own playing. Every stem is independent, so the remix is yours to shape.
Working on a live recording or fancam instead? KROMA also detects crowd, applause and speech as separate tracks — clean the audience out while you are mixing the stems.
Segment processing for long songs
Only need the instrumental for a chorus or a specific section? Set a time segment and KROMA processes just that part — the rest of the song passes through untouched, saving both time and credits.
Questions
How clean is the instrumental?
The vocal stem is removed from the mix, not filtered out — the band keeps its low-end weight and stereo image. On dense wall-of-sound mixes a faint vocal ghost may remain, but it sits far below karaoke-usable level.
Can I keep a little of the vocal as a guide?
Yes — the vocal fader is not just on/off. −12 to −18 dB leaves a quiet guide vocal that's handy for practice and cue points.
Does it work on live recordings?
Yes — and the analysis can additionally detect crowd, applause and speech as separate tracks, so you can clean the audience out of a concert recording while making the instrumental.
Can I get the acapella instead of the instrumental?
Yes — invert the faders: mute everything except vocals. There's also a dedicated acapella mode tuned for the cleanest possible vocal extraction.
Free to try — no install, works in the browser.