Part of our roundup of the best sheet music scanner apps.
The short answer: optical music recognition, or OMR, is software that reads sheet music from a photo or PDF and turns it into a score a computer understands, so you can play it back, edit it, transpose it and export it as MusicXML or MIDI. It’s OCR for music. Songscription’s sheet music scanner runs on AnnieScan, our own OMR model, and opens every scan in Songscription’s score editor.
Songscription is our product. We put it first where it’s the right tool, and we say plainly where another tool fits better.
What OMR does
A scan or PDF of sheet music is a picture. You can read it, but a computer sees only pixels. OMR finds the music in those pixels: staves, clefs, key and time signatures, note heads, stems, beams, rests, accidentals, ties, slurs, dynamics, lyrics and chord symbols. Then it works out what they mean together, which pitch each note is, how long it lasts, which voice and which part it belongs to, and writes all of that as a digital score.
The output is notation, not sound and not an image. That’s what lets you hear the page, change the key, or open it in a notation app without typing every note again.
Why reading music is harder than reading text
Text OCR reads one line of letters, left to right. Music doesn’t behave like that.
- It’s two-dimensional. Chords stack notes vertically, and two or more voices share one staff.
- Meaning sits far away. The key signature at the start of a line changes every note after it. A tie can cross a barline, a slur can cross a line break, and a dynamic sits under the staff, away from its notes.
- Symbols touch and overlap. Beams join note groups, ledger lines cross stems, and accidentals crowd against note heads.
- Small marks matter. A missed dot changes a rhythm; a missed accidental changes a pitch for the rest of the bar.
- Every engraver is different, and handwriting differs again.
That’s why a good scanner has to understand music, not only shapes.
How modern OMR works
Older OMR followed hand-written rules: find the staff lines, remove them, then match what was left against templates of symbols. It worked on clean print and broke on everything else.
Modern OMR uses machine learning. Models trained on large collections of scores learn to read the page as a whole, the way a musician does, and handle the photos, handwriting and crowded pages that rule-based systems couldn’t. That’s the approach behind AnnieScan, and behind most current research.
Where OMR research is heading
OMR is an active research field. Projects such as OmniOMR, led by Jan Hajič jr. at Charles University in Prague and funded by the Czech Ministry of Culture from 2023 to 2027, use machine learning to read the historical music collections held in Czech libraries, so that musical heritage can be cataloged, searched and opened up. Work like this, published as open research and open-source code, keeps pushing what OMR can read. It isn’t an app you scan your own music with; for that, you need a scanner like Songscription’s.
What you can do with OMR today
- Edit a printed score without retyping it: fix a part, add a verse, rearrange it.
- Transpose a song for a different voice or instrument.
- Hear a piece before you learn it, slowed down or looped.
- Move music between apps as MusicXML: from a paper part into MuseScore, Sibelius or Dorico. Our guide to converting a PDF to MuseScore shows how.
- Digitize a library of old parts, method books or your own handwritten compositions.
OMR in Songscription
Songscription’s sheet music scanner reads a PDF or phone photos, up to 50 pages in one scan. AnnieScan Mini is fast on clean print. AnnieScan Studio, our most accurate model, also reads handwritten scores, chord symbols, lyrics and tabs, with the dynamics and tempo words.
The scan opens in the score editor, where you play it back, slow it down, change the key and fix any note, or ask Annie, the AI score assistant, to make the change. Then export MusicXML, MIDI, PDF or Guitar Pro on Plus and Pro. Free accounts scan a page a day.
OMR vs transcription
OMR reads notation that already exists. Transcription listens to a recording and writes notation that didn’t exist before. If you have the page, scan it; if you have only the song, transcribe the audio. Our guide to audio transcription vs sheet music scanning goes through which one you need, and our roundup of the best sheet music scanner apps compares the scanners.
How we checked
We read the OmniOMR project page in November 2026.
Sources:
Frequently asked questions
What is optical music recognition?
Optical music recognition, or OMR, is software that reads sheet music from an image or PDF and turns it into an editable digital score, such as MusicXML or MIDI. It’s the music equivalent of optical character recognition for text. Songscription’s sheet music scanner uses AnnieScan, our own OMR model, and opens the result in Songscription’s score editor.
Is OMR the same as OCR?
The idea is the same, reading symbols from an image, but music is much harder. Text runs in one line, left to right. Music stacks notes into chords, runs several voices on one staff, and spreads meaning across symbols that sit apart from each other, such as a key signature at the start of the line, a tie across a barline or a dynamic below the staff.
How accurate is optical music recognition?
It depends on the page. Clean, printed engraving reads best. Low-resolution scans, phone photos taken at an angle, dense orchestral scores and handwriting are harder for every OMR tool, so always check the result, especially accidentals and rhythms. In Songscription you fix any note in the score editor after the scan.
Can OMR read handwritten music?
Some OMR tools can. Songscription’s AnnieScan Studio reads handwritten scores, along with chord symbols, lyrics and tabs. Many other scanners read printed music only.
What is the difference between OMR and music transcription?
OMR reads notation that already exists on a page. Music transcription listens to a recording and writes notation that didn’t exist before. Songscription does both, and both end in the same editable score.
