A full band, a horn section, or a string quartet is the hardest thing to transcribe, because several instruments compete for the same space in one recording. For a long time the honest advice was to work one instrument at a time. You still can, and now Songscription’s Multi-Instrument mode can also transcribe every instrument in one pass and hand back a part for each. In our October 2026 benchmark on multi-instrument recordings, Songscription’s sheet music scored 61.1 to Klangio’s 46.5 and its MIDI 52.2 to 40.4, first against every tool we tested.
Songscription is our product. We put it first where it’s the right tool, and we say plainly where another tool fits better.
Multi-instrument guides by goal
If you want to know what AI can and can’t do with several instruments, start with can AI transcribe multiple instruments at once? To choose between one pass and one part at a time, read one-pass vs per-instrument transcription. How to transcribe a full band recording takes you through a band from upload to parts, and transcribe multi-track audio to sheet music is for working from stems or multi-track audio.
When you’re ready to hear what it does with your own music, upload a recording and try Songscription free, or create a free account first.
Three ways to transcribe several instruments
The first route gets you every part at once. Choose Multi-Instrument when you upload, and Songscription transcribes the whole recording in one pass. You get a separate part for each instrument it hears, from drums and bass to strings, brass, and saxophones, as a multi-staff score and as multitrack MIDI. Keyboard parts are split into left and right hands, so a piano or synth part reads like real keyboard music. Band leaders handing out parts, arrangers writing out a horn or string section, and producers rebuilding a song in their DAW usually want this one.
The second route gets one part as clean as possible. Pick a single instrument, such as bass guitar or vocals, and the model listens only for that sound and ignores the rest of the mix. With one job to do, it’s still the most accurate way to get a single part, so use it when one line matters more than the others.
The third gives you one playable version of the whole song. If you want to play the song yourself and don’t need every instrument written out, a piano cover, a piano + vocals arrangement, or a lead sheet condenses the arrangement into one part that carries the melody and harmony.
What Multi-Instrument can transcribe
Multi-Instrument recognizes nine instrument families and writes each instrument it hears as its own track.
| Family | What it includes |
|---|---|
| Keys | Acoustic piano, electric piano, organ, and mallet percussion such as vibraphone, marimba, and glockenspiel |
| Guitar | Acoustic, clean electric, and distorted electric guitar |
| Bass | Upright bass and electric bass |
| Strings | Violin, viola, cello, contrabass, harp, timpani, string ensemble, and synth strings |
| Brass | Trumpet, trombone, tuba, French horn, and brass section |
| Reeds and winds | Soprano, alto, tenor, and baritone sax, oboe, English horn, bassoon, clarinet, flute, piccolo, and recorder |
| Synth | Synth lead and synth pad |
| Voice | Sung parts, as a voice track |
| Drums | Drum kit |
Keyboard tracks are split into left and right hands. A choir comes back as one voice track, without separate soprano, alto, tenor, and bass parts. The more instruments overlap in the same range, as in a dense orchestral recording, the more cleanup to expect. When one part matters most, single-instrument mode is still the cleanest way to get it.
One pass or one part at a time?
Multi-Instrument gives up a little accuracy per part in exchange for everything from one upload, and single-instrument mode makes the opposite trade. On a dense mix, some parts will need more cleanup than others. A pattern that works well is to run Multi-Instrument for the whole song, then rerun the part you care about most in single-instrument mode and swap it in. One-pass vs per-instrument transcription has the full comparison, and can AI transcribe multiple instruments at once covers the broader question of what AI can do with several instruments.
Do you need stems?
No. Multi-Instrument and single-instrument mode both work from the full mix. Stems, the separated instrument tracks of a song, can help, and you don’t need them. If a song already exists as separate tracks, transcribing each stem by itself gives the cleanest parts of all, because no model has to pull anything out of a mix.
A concert or rehearsal tape caught on one room mic is the hardest version of this. Our roundup of the tools that cope best with live recordings covers what to expect and which optional cleanups help. What is stem separation explains what stems are and when they’re worth it, and transcribing multi-track audio to sheet music walks through the multi-track workflow end to end.
From parts to a finished score
With Multi-Instrument, the parts arrive already assembled as one score in the score editor we built. Review each part and fix anything the model missed, by hand or by asking Annie, the AI score assistant. Then download the score as PDF or MusicXML, or the multitrack MIDI for your DAW. Every instrument also gets its own linked part, so you can print a part for each player. For the step-by-step from upload to finished parts, see how to get sheet music for every instrument in a song.
If you built the parts one instrument at a time instead, you combine them into a multi-staff score yourself, and how to build a full score from separate parts walks through it. For getting each individual part right, the per-instrument techniques for piano, guitar, bass, horns, strings, voice, and drums are in the instrument transcription guide.
Related guides
For the approach, read can AI transcribe multiple instruments and one-pass vs per-instrument. The workflow is in transcribe a full band recording and transcribe multi-track audio. Stem separation and the instrument transcription guide cover the pieces, how to separate stems before transcribing goes step by step, and how to build a full score from separate parts covers putting the score together.
Frequently asked questions
Can AI transcribe a full band at once?
Yes. Songscription’s Multi-Instrument mode transcribes a mixed recording in one pass and gives you a separate part for each instrument it hears, including keys, guitars, bass, strings, brass, woodwinds, saxophones, synths, voice, and drums, as sheet music and as multitrack MIDI. Expect to check each part and fix the odd note in the editor. If you only need one part and want it as clean as possible, pick that instrument by itself instead, and the model isolates it directly from the mix.
Should I use Multi-Instrument or transcribe one instrument at a time?
Use Multi-Instrument when you want every part of a song from one upload, for example to hand out parts or rebuild the song in your DAW. Pick a single instrument when one part matters most, such as a bass line you want to learn note for note, because the model can put all its attention on that one sound. Plenty of people do both, running Multi-Instrument for the whole picture and a single-instrument pass on the part they care about most.
How do I get a full score from a band recording?
Upload the recording, choose Multi-Instrument, and Songscription returns a multi-staff score with a part for each instrument it hears, plus multitrack MIDI. Keyboard parts come split into left and right hands. Review the parts in Songscription’s own score editor, then download PDF, MusicXML, or MIDI. If you only need one playable version of the song, a piano cover or lead sheet is faster than a full score.
What are stems, and do I need them?
Stems are the separated instrument tracks of a song, such as the drums, the bass, and the vocals, split out from the full mix. You don’t need them. Multi-Instrument works from the full mix, and so does single-instrument mode, which isolates the part you pick directly. If a song already exists as separate tracks, transcribing each stem by itself gives the cleanest parts of all.
The fastest way to start is to upload a song you want every part of. Try Multi-Instrument on your recording.
