Guide
How to separate vocals and instruments from a song
Until recently you needed the original multitrack session to hear an instrument on its own. Today, AI stem separation splits a finished MP3 into its parts in about a minute. Here's what it is and how musicians use it.
What a "stem" is
A stem is one part of a mix: all the vocals, all the drums, the bass, and so on. Separation models typically output:
- Vocals (lead and backing together, sometimes split)
- Drums
- Bass
- Guitar
- Piano / keys
- Other (strings, synths, horns, effects)
How it works
Separation models are neural networks trained on thousands of songs where the individual tracks are known. They learn what a voice, a kick drum or a bass guitar "looks like" in a spectrogram and build a mask for each one. The best current models (transformer-based, such as the RoFormer family) produce vocals clean enough for karaoke and transcription.
How good is it?
| Stem | Typical quality |
|---|---|
| Vocals | Very good; a little reverb or bleed on dense mixes |
| Drums | Very good |
| Bass | Good; sometimes loses the lowest notes or picks up kick drum |
| Piano / guitar | Fair to good; they overlap in pitch and timbre |
| Other | Variable |
Quality drops on live recordings, heavily distorted mixes and old mono recordings.
What musicians use stems for
- Practice: mute your own instrument and play along with the rest of the band.
- Learning parts: solo the bass or the backing vocals to hear exactly what they play.
- Transcription: a melody or bass model works far better on an isolated stem than on the full mix.
- Arranging: hear the original voicings, fills and counter-lines before writing your own.
- Remixes and mashups (for private use, or with the rights cleared).
A note on copyright
Separating a song for your own practice, study and arranging is everyday musicianship. Publishing the separated tracks or using them in a release needs the rights holder's permission.
Stems in Medley Writer
Every song you upload is split into vocals, drums, bass, guitar and keys. You can solo or mute stems while listening, and each part of the generated score is transcribed from its own stem, so the bass line comes from the bass, not from a guess at the full mix.
Let Medley Writer do the transcribing
Upload the songs, choose the parts and the order, and get a full arrangement for your band with the transitions written in. Free during the beta.
Join the waitlist