Part of our guide to transcribing a full band.
Stems are separated audio: the vocals, drums, bass, and other parts of a song split into their own recordings, so you can mute, solo, and remix them. Multitrack MIDI is the notes each instrument plays, one track per instrument, with no sound attached, so you can re-voice, edit, and arrange them. If you want to keep the original sound, you need stems. If you want the notes, you need MIDI.
Both get described as "splitting a song into instruments," which is why people mix them up. This guide covers what each one actually contains, what each is good for, and how to get the one you need.
What are stems?
Stems are audio files, each holding one part of a song. A typical split gives you vocals, drums, bass, and "other" (everything else), and some tools split further into piano or guitar. Play all the stems together and you hear the original song.
Stems come from two places. Sometimes they come from the original session, where each instrument was recorded on its own track. More often they come from a stem-separation tool, which uses an AI model to pull each part back out of a finished mix. Separated stems are not perfect: you can hear faint bleed from other instruments and some watery artifacts, especially on dense songs. Our explainer on what stem separation is covers how it works and where it struggles.
The key point is that a stem is still sound. It carries the singer's actual voice, the real snare, the exact tone of the guitar amp. You can change its volume, add effects, or cut it up, but you cannot easily change which notes it plays.
What is multitrack MIDI?
Multitrack MIDI is a MIDI file with a separate track for each instrument, where every track is a list of notes: which pitch, when it starts, how long it lasts, and how hard it is played. There is no audio in it at all. Load it into a DAW and each track plays through whatever instrument you assign, a piano, a synth, a drum kit, a sampled string patch.
Because it is note data, every musical detail is editable. Move a note, transpose the whole song, change a chord voicing, slow it down without any artifacts, or give the guitar part to a Rhodes. If you are new to the format, what is MIDI is a good primer.
Multitrack MIDI is also what you get when a recording is transcribed part by part. Songscription's Multi-Instrument mode (beta) produces it in one pass, with a track for each instrument it hears, from drums and bass to strings, brass, and synths, alongside a multi-staff score.
Stems vs multitrack MIDI side by side
| Stems | Multitrack MIDI | |
|---|---|---|
| What it is | Separated audio of each part | The notes of each part, as data |
| Sounds like | The original recording | Whatever instrument you assign |
| Edit individual notes | Hard; you are editing audio | Easy; drag, delete, transpose |
| Change the instrument | No | Yes, any virtual instrument |
| Change key or tempo | Possible with processing, can add artifacts | Clean, no artifacts |
| Turn into sheet music | Needs transcribing first | Can be shown as notation |
| Typical artifacts | Bleed and smearing from separation | Wrong or missing notes from transcription |
| Best for | Remixes, karaoke, practice tracks, sampling | Re-voicing, arranging, learning, rebuilding a song |
| How to get it | Original session files or a stem-separation tool | Transcription, or recording MIDI as you play |
When do you need stems?
Reach for stems when the original sound is the point:
- Karaoke and backing tracks. Remove the vocal and sing over the real band.
- Practice tracks. Mute the bass to play along with the rest of the band, or solo the bass to hear exactly what the player did.
- Remixes and mashups. Build a new track around the real vocal take.
- Sampling. Chop a drum break or a horn stab as audio.
None of these care about note names. They need the recording itself, just in pieces.
When do you need multitrack MIDI?
Reach for multitrack MIDI when the notes are the point:
- Re-voicing a part. Play the original bass line through your own analog patch, or move a guitar riff onto a synth. Our guide to how producers use audio-to-MIDI to recreate sounds covers this one-part move in detail.
- Rebuilding a whole song. Load every part into your DAW and produce your own version. The full workflow is in how to recreate a song in your DAW.
- Arranging. Take the parts, change the key, rewrite the harmony, or adapt them for a different lineup.
- Learning. See exactly what each instrument plays, on a piano roll or as sheet music.
A quick test: if you would be happy hearing the part on a different instrument, you want MIDI. If you need the singer's real voice or the drummer's real kit, you want stems.
How do you get each one?
To get stems, use the original multitrack session if you have it, or run the mix through a stem-separation tool. If you made the song in Suno, our guide on getting stems from a Suno song covers that case. Songscription does not export audio stems.
To get multitrack MIDI, transcribe the recording. In Songscription, upload a recording or paste a YouTube, TikTok, or Instagram link, choose Multi-Instrument, and download the MIDI. The Download menu offers two MIDI flavours: the score-derived MIDI, which matches the cleaned-up sheet music and includes your edits, and the raw transcription MIDI, which keeps the free timing and velocities of the performance. Because Multi-Instrument is in beta, expect to fix some notes, and see cleaning up MIDI after conversion for the usual fixes.
Can you use both together?
Yes, and they combine well. Stems make cleaner MIDI: if one part is buried in a dense mix, split it into a stem first, then transcribe that stem in single-instrument mode for a more accurate track. Our walkthrough on separating stems before transcribing covers that step. Going the other way, producers often keep the original vocal stem and rebuild everything underneath it from MIDI.
You do not need stems to get multitrack MIDI, though. Multi-Instrument works from the full mix.
Related guides
- The hub: the multi-instrument transcription guide.
- About stems: what is stem separation and separating stems before transcribing.
- About MIDI: what is MIDI and cleaning up MIDI after conversion.
- Put it to work: recreate a song in your DAW.
Frequently Asked Questions
What is the difference between stems and multitrack MIDI?
Stems are audio. Each stem is a recording of one part of the song, such as the vocals, drums, or bass, split out from the mix, and it sounds exactly like the original. Multitrack MIDI is note data. Each track lists the notes one instrument plays, when, and how hard, with no sound attached. Stems let you hear and remix the original performance. MIDI lets you change the notes and play them through any instrument you choose.
Should I use stems or MIDI for remixing?
Use stems when you want to keep the original sound, for example to make a karaoke track, an a cappella, a practice track without the bass, or a remix built on the real vocal. Use MIDI when you want the notes but not the sound, for example to play the chord progression on your own synth, rebuild the bass line with a different patch, or arrange the song for other instruments.
Can I turn stems into MIDI?
Yes. A stem is just audio of one instrument, so you can transcribe it like any other recording. Transcribing a clean stem in single-instrument mode is one of the most accurate ways to get MIDI for that part, because nothing else in the mix competes with it. You can also skip stems entirely and transcribe a recording into multitrack MIDI in one pass.
Does Songscription export audio stems?
No. Songscription transcribes audio into notes, so it gives you sheet music, MIDI, MusicXML, and, for guitar, Guitar Pro files, not separated audio. Multi-Instrument mode (beta) returns multitrack MIDI with a separate track for each instrument it hears. If you need the separated audio itself, use a stem-separation tool alongside it.
Want the notes of every instrument in a song? Convert a recording to multitrack MIDI.
