What OMR does
Optical music recognition (OMR) converts a picture of sheet music — a scan, a photo or a PDF — into a structured, machine-readable score, usually MusicXML. Once the music is data rather than pixels, software can play it, transpose it, reformat it or analyse it.
It is harder than text OCR. Musical meaning depends on position (how high a note sits on the staff), on context (a sharp in the key signature changes every later F) and on relationships between symbols (beams, ties, slurs and voices that cross each other). A good OMR engine has to understand all of that, not just recognise shapes.
How an OMR engine works, step by step
- Image clean-up — the image is converted to black and white, de-skewed and cleaned of specks.
- Staff detection — the engine finds the five-line staves, measures their spacing (which sets the scale for everything else) and groups them into systems.
- Symbol detection — noteheads, stems, beams, flags, rests, clefs, accidentals, dots, bar lines and time signatures are located and classified.
- Assembly — symbols are combined into notes and chords with durations, voices and measures, and checked against rules (a 4/4 bar should add up to four beats).
- Export — the result is written out as MusicXML.
The engine behind Musaica
Musaica is a free online sheet music reader built on Audiveris, the leading open-source OMR engine. Audiveris normally runs as a desktop Java application; Musaica runs it on its servers so you can use it from any browser, then engraves the result with OpenSheetMusicDisplay and plays it back. See Audiveris online for when the desktop app is the better choice.
Where OMR is reliable — and where it is not
| Usually reliable | Often needs checking |
|---|---|
| Notes, rests, beams, dots, ties in printed music | Grace notes and ornaments |
| Clefs, key and time signatures at the start of a line | Clef or key changes in the middle of a bar |
| Single-line melodies and piano music | Dense orchestral scores and cross-staff beaming |
| Clean digital PDFs and 300 DPI scans | Handwriting, faint photocopies, curved or skewed photos |
| Standard accidentals | Tuplets, lyrics, chord symbols, microtonal (maqam) accidentals |
Image quality makes the biggest difference. Our scanning guide shows how to prepare pages so OMR gets them right first time.
How to try OMR online with Musaica
- Upload a score. A PDF, PNG or JPG of printed notation.
- Recognition runs on the server. Audiveris analyses each page and produces MusicXML.
- See and hear the result. The score is engraved in your browser and plays back with a cursor, so errors are easy to spot by ear.
Frequently asked questions
What is optical music recognition?
Optical music recognition (OMR) is software that reads images of sheet music and converts the notation into a digital score format such as MusicXML.
Is OMR the same as OCR?
It is the musical equivalent. OCR recognises letters; OMR recognises staves, notes, rests, clefs and other musical symbols and how they relate.
What is the best free OMR software?
Audiveris is the leading free, open-source OMR engine for desktop. Musaica runs Audiveris online so you can use it without installing anything, and adds playback.
How accurate is OMR?
On clean printed music, modern OMR gets the large majority of notes right. Accuracy drops with handwriting, poor scans and complex notation such as grace notes and tuplets.
Can OMR read handwritten music?
Mostly not yet. Handwritten music recognition is an active research area; current tools work best on printed engravings.