A stem separator takes a finished mix and gives you back the parts: vocals, drums, bass, guitar, piano and everything else, each on its own fader. Mute the part you play, solo the one you are learning, slow the song down, and export the result as a WAV. In the browser, free, no account.
The model was trained on thousands of real multitrack recordings, so it has learned what a bass sounds like against a kick drum and what a voice sounds like against a guitar. Given a mix it estimates each part and writes it out as its own file. Nothing is looked up or matched; your song is processed from scratch.
Most separators stop at vocals, drums, bass and other. Here guitar and piano are separated as well, which matters if one of those is the instrument you play: you can take your own part out and keep everything else.
Vocals, drums and bass come out cleanest, because they occupy their own space in a mix. Guitar and piano are harder, especially when they play the same chords, and a little of one may bleed into the other. Dense, heavily compressed mixes are the hardest; sparse recordings separate best.
The stems are rendered at your song's own sample rate and exported as 24-bit WAV, with no re-compression and no watermark. What you hear in the mixer is what ends up in the file.
Instead of a zip of files to carry into a DAW, the stems land on faders with mute, solo, low and high shelf, pan and a loop. Slow the song down without changing the pitch and practise the part before you export anything.
Learning a bass line or a solo by ear, playing with the rest of the band, making a karaoke or backing track, hearing how a mix is built one instrument at a time, or getting a drumless version to practise drums to.
MP3, WAV, FLAC or M4A, the whole song. It uploads to the separation server and nothing else is asked of you.
The separation runs on a GPU in the cloud, about a minute for a typical song. Afterwards the song and its stems stay in your browser's own storage.
Pull your part down, solo the one you are learning, loop a section. Export a 24-bit WAV mix when it is right, or send it to CountIn for a count-in.
A tool that takes a finished stereo mix and splits it back into its parts, called stems: vocals, drums, bass and so on. Modern separators use a neural network trained on multitrack recordings; older ones relied on phase tricks that only removed centred vocals.
Good enough to practise with and to make backing tracks from. Vocals, drums and bass are usually clean. Guitar and piano can bleed into each other on busy mixes, and a solo stem played on its own will show some artefacts that disappear once the other parts are back in. It is not a substitute for the original multitracks.
Yes: two separations a day per person, full-length songs, 24-bit WAV export, no account and no watermark. The cap resets every day.
Six: vocals, drums, bass, guitar, piano and other, which holds everything the model could not assign to the first five, like synths and strings.
You export the mix you have set up in the mixer as a WAV, so muting everything but one part gives you that stem. The stems themselves are also kept in your browser for the next time you open the song.
No. The separation runs in the cloud and the mixer runs in the browser tab, so it works on a laptop, a tablet or a phone.
When your mix is ready for a gig, send the file to CountIn, our sibling tool. It bakes a synced count-in into the track, so the band comes in together on beat one from any phone or playback app, and it can transpose the song to your singer's key without changing the tempo.
Add a count-in with CountInDrop in a song and have your stems in a few minutes.
Start Free, No Sign Up