Stem Separation
A Mixer That Shows Only What's Actually There
Most stem splitters hand you the same fixed four faders whether they're in the song or not. KROMA analyzes your track first — then lists only the sounds it really found, each with its own fader and mute switch.
How it works
- 1
Drop in your track
WAV or MP3. Studio mixes, live recordings, phone captures — anything works.
- 2
Analyze the sounds
Neural models scan the whole file and report which sounds are present — instruments, vocals, crowd, wind — with coverage showing where each appears.
- 3
Mute or rebalance each sound
Every detected sound gets a fader from −24 to +12 dB plus a mute. Silence the drums, pull the crowd to zero, push the vocal forward.
- 4
Download your mix
The stems are separated, your gains applied, and the result re-summed into a single WAV.
Sounds KROMA can isolate
Beyond the four-stem wall
Standard stem splitters stop at vocals / drums / bass / other — and everything that isn't a Western pop instrument collapses into that anonymous "other" bucket.
KROMA goes deeper. A second analysis layer scans the residual for individual instruments — strings, woodwind, brass, synths, and Chinese instruments like erhu, guzheng, dizi, suona and sheng that conventional detectors can't name at all. If the song has it, you get a fader for it.
Non-musical sounds are tracks too
Crowd, applause, speech, wind, rain, traffic, sirens, electrical hum — the analysis treats the recording's environment the same as its instruments. That's what makes the mixer work for live recordings and field captures, not just studio tracks: the noise floor is a fader, not a fate.
Honest about what's detectable
The mixer only lists sounds the analysis actually found — if there are no bagpipes in your track, there is no bagpipe fader. Coverage percentages show where each sound appears, so you can trust the list instead of guessing.
Questions
Can I remove one instrument from a song?
Yes — analyze the track, then mute the fader for that instrument. KROMA detects beyond the usual four stems: individual instruments like violin, saxophone, erhu or guzheng get their own fader when present.
Will muting a sound leave artifacts?
Core stems (vocals, drums, bass, guitar, piano) separate cleanly. Individually extracted instruments and environmental sounds may leave slight residue when fully muted — the gain slider often gives a more natural result than a hard mute.
Can it detect Chinese instruments?
Yes — erhu, pipa, guzheng, yangqin, dizi/xiao, suona and sheng are detected via a zero-shot audio model, since standard classifiers don't have categories for them.
Can I process only part of the track?
Yes — select a time range before rendering and only that segment is processed; the rest passes through unchanged.
Free to try — no install, works in the browser.