If you’ve poked around music tools lately, you’ve probably seen the word stems. It sounds technical, but the idea is simple.
What a ‘stem’ is
A finished song is a blend of many parts mixed together — vocals, drums, bass, guitars, keys and more. A stem is one of those parts on its own. The vocal stem is just the singing. The drum stem is just the drums. Separating a song means taking the final mixed track and pulling those parts back out.
Why it used to be impossible
Once a song is mixed and exported, all those parts are blended into two channels (left and right). It’s like pouring several cups of coloured water into one jug and then being asked to pour them back into separate cups. For decades, that was considered impossible to do well.
How AI changed it
Modern separation uses a neural network trained on enormous amounts of music where the original separate parts were known. Over time it learned the ‘fingerprint’ of a singing voice, a snare drum, a bass guitar and so on — across frequency and time. When you feed it a new song, it predicts which sound belongs to which instrument and rebuilds each part.
Vocal Studio uses a well-known open model in this family. The clever bit is that it runs this model inside your browser, on your own device — so you get studio-grade separation for free, and your audio is never uploaded anywhere.
What you can get
- Vocals — the isolated singing.
- Instrumental — everything except the lead vocal.
- Drums, Bass and ‘Other’ — with the full split. ‘Other’ is everything that isn’t drums, bass or vocals: guitars, pianos, synths and so on.
It isn’t magic and it isn’t perfect, but it’s good enough that musicians, worship teams, DJs and learners use it every day.