§ AI MUSIC TRANSCRIPTION · NOTATION

Turn a recording into sheet music

The hard part of transcribing by ear is not hearing the notes — it is writing them down at the right rhythmic value against the right bar line. That is the part automatic music transcription has to get right, and the part most tools skip.

Notes are the easy half

Pitch detection is a solved-enough problem for most material. What separates a usable transcription from a pile of note events is metrical interpretation: is that a dotted eighth or a triplet, does the phrase start on the beat or before it, and where exactly is bar one.

Getting that wrong produces notation that is technically correct and completely unreadable. So tempo, downbeat and meter are detected first, and the notes are fitted to that grid rather than the grid being inferred afterwards.

Rubato is not an error

Real performances breathe. A pianist slows into a cadence; a singer sits behind the beat. A transcriber that assumes a fixed tempo will smear all of that into wrong rhythms.

The tempo map here is allowed to move, so an expressive performance is transcribed against the tempo it was actually played at rather than being forced onto a click.

A score, not a stack of staves

Transcribe a band and you get a real system: every part bracketed together, bar lines connected, and bar three of the bass sitting directly under bar three of the horns. A part that stops early keeps counting rests, the way it would on paper, instead of leaving the system ragged.

Each part is engraved on the clef that actually suits it. A bass line lands on bass clef rather than hanging three ledger lines under a treble staff, and a viola part gets alto clef.

Key signatures, not a wall of accidentals

Most audio-to-notation output gives away its origin immediately: a sharp printed on every single F and C, because the tool only knows MIDI numbers and not keys. That is not how music is written.

The key is detected from the pitch content, put in the key signature, and only the notes that depart from it are marked — with accidentals persisting across the bar, as notation actually works. The spelling follows the key too: the same black key is written D flat in A flat major and C sharp in D major.

Transposing instruments are written correctly

Transcription hears concert pitch. A trumpet player does not read concert pitch. A B flat trumpet part is read a major second above what it sounds, an E flat alto sax a major sixth, an F horn a fifth — and the key signature moves with the notes, so a concert E flat major chart is written in F major on the trumpet stave.

Get this wrong and the part is unplayable by the person you handed it to. Each instrument carries its own transposition, and a concert-pitch toggle collapses everything back to sounding pitch when you want the conductor view.

Fix it where you see it

No transcription of a real recording is perfect. The same result opens as an editable multitrack MIDI arrangement — one lane per part — where you correct a wrong note, and the score re-engraves from that same edit. The MIDI you export is what you fixed, not what the model first guessed.

Where it goes next

The result is note data with a tempo map and chord symbols, which is what a notation program or a DAW both want. Export to a Logic or Ableton session keeps the tempo map intact, so the notes stay where the performance put them.

Transcribe a recording

Drop an audio file and get notes, a score and MIDI back. Free runs daily.

Open the transcriber →

FAQ

What is automatic music transcription?

Automatic music transcription (AMT) is turning recorded audio back into symbolic music — notes, rhythms, bars and chords — rather than into text. An AI transcription model detects the pitches and onsets, and the notation layer decides how to write them: rhythmic values, bar lines, clefs and key signature.

Will AI transcription produce publication-ready engraving?

It produces a readable score — correct notes, rhythms against a real tempo map, per-part clefs, a key signature and written transposition. Final editorial engraving (spacing, cue sizing, page turns, articulation polish) still belongs in a dedicated notation editor, and the export is designed to hand off cleanly to one.

Does it transcribe multiple instruments at once?

Yes. The mix is separated first and each part is transcribed on its own, then engraved as its own stave in a bracketed system with the bar lines aligned — a score you read down, not one collapsed line.

Does it handle transposing instruments like B flat trumpet?

Yes. Notes and key signature are transposed together, so a concert E flat major part is written in F major for the trumpet player, and a concert-pitch toggle puts every part back at sounding pitch.

What about chord symbols?

Chords are detected as part of the same pass, including over a moving tempo, and come through as a chord track alongside the notes.