If you have ever wanted to sing along to a song without the original singer, use only the beat of a track in a reel, or turn the drums down because they drown the vocal, you have wanted stems. The word sounds technical, but the idea is simple.
This guide explains what a stem is, how it differs from a track or a multitrack, how AI pulls stems out of a finished song, and what people actually use them for.
A stem, in plain words
A finished song is many sounds mixed together into one stereo file. A stem is one group of those sounds saved separately. A drums stem contains the whole kit: kick, snare, hi-hats and cymbals, already balanced. A vocals stem contains the lead singer and often the backing vocals. A bass stem contains only the bass.
The useful property of stems is that if you play all of them together at their original level, you get the full song back. That is what makes them easy to work with: you can change one part without touching the others.
Stems, tracks, multitracks and acapellas
These words are often mixed up, so it helps to separate them.
- Track: in a recording session, one individual recording, such as a single microphone on the snare drum.
- Multitrack: the full set of individual tracks from a session, often dozens of them.
- Stem: a group of related tracks combined into one file, such as all the drums or all the vocals. A song might have four to eight stems.
- Acapella: just the vocals, which is the vocal stem on its own.
- Instrumental or karaoke track: everything except the lead vocal.
Why stems exist
In professional music, stems are delivered for practical reasons. A mixing engineer may receive stems to balance a song. A film or TV editor receives stems so the music can be ducked under dialogue without losing the beat. DJs and remixers ask for stems to build new versions. Live acts often play backing stems alongside real musicians on stage.
Producers who made the song have the stems already. Everyone else, until recently, had only the finished stereo file and no clean way to get the parts back.
How AI separates a finished song
Stem separation, also called source separation, is the AI technique that recovers stems from a finished mix. In general terms, a model is trained on many songs where both the full mix and the original stems are known. It learns what a voice, a drum hit or a bass note typically looks like in the frequency content of audio, and learns to estimate which parts of a new mix belong to which source.
It is an estimate, not a perfect undo. You may hear a faint trace of cymbals in the vocal stem, a slightly hollow sound where instruments overlapped, or reverb split unevenly. Results are usually cleanest on songs with a clear mix and fewer overlapping instruments.
What people use stems for
In India, a few uses come up again and again.
- Karaoke for a sangeet or family function: mute the vocal stem and sing over the rest.
- Riyaz and learning: mute the guitar or piano stem and play that part yourself.
- Reels and edits: use only the beat under a voiceover, or only the vocal over new visuals.
- Remixes: build a dance version of a bhajan or a lofi version of a romantic song from the parts.
- Fixing balance: turn down an instrument that crowds the singer.
Stems on Autunes
On Autunes, Stems splits a song into six parts: vocals, drums, bass, guitar, piano and other. Each comes as a 256 kbps MP3. It costs 5 credits a song, is available on Pro and Creator, takes about a minute for a four-minute song, and your credits are refunded if the split fails. You can split songs you made on Autunes or tracks you upload in MP3, WAV, FLAC, M4A or OGG, up to 50 MB and 10 minutes, after confirming you own the audio or have the rights.
Instruments that are not one of the five named groups, such as flute, strings, sitar or synth pads, generally land in the other stem, and keyboard-like sounds can be shared between piano and other. Creator also includes extracting any instrument, and both paid plans include MIDI export if you want the notes rather than the audio.
A note on rights
Separating a song does not make it yours. Making a karaoke version of a film song for practice at home is one thing; posting it, performing it publicly for money or remixing it for release needs permission from whoever owns the rights. Songs you made yourself, on Autunes or elsewhere, are the safest material to work with.