What is Stem Separation?
Also called: source separation, audio separation, demixing
Stem separation is the process of splitting a finished, mixed song into its parts, such as vocals, drums, bass and other instruments, using AI models trained to recognize each source.
A mixed song is one waveform, and the instruments overlap in time and frequency, so separating them is a hard problem. Modern models learn what each kind of source sounds like from large amounts of music and estimate each part, which works far better than older tricks based on EQ or phase.
The results are not perfect copies of the original recordings. Expect some bleed, where a little of one instrument remains in another stem, and occasional artifacts on reverb tails, cymbals or heavily processed sounds.
Common uses are making instrumentals and acapellas, practicing along with a band without one instrument, studying a mix, remixing and sampling. Using separated parts in a release still requires the rights to the original.
Put it to work
Separate any song into stems with AI.
Stem Isolator