Converting PDF Sheet Music to MusicXML: Tips for Clean Results
Optical music recognition explained: how to prepare scans, what usually goes wrong, and how to check the result.
What optical music recognition does
Optical music recognition (OMR) is to sheet music what text recognition is to books: software finds the staves, clefs, notes, rests and symbols in an image and rebuilds them as structured data, usually MusicXML. Once a score is MusicXML, it can be played back, transposed, edited, or shown as falling notes.
Piano: Soloist uses an open-source OMR engine on its server to convert PDF pages, then opens the result directly in the player.
Preparing a good PDF
Recognition works best on clean, printed scores. Use a scan of at least 300 dpi, keep pages straight, and avoid shadows, handwriting and highlighter marks. Digital PDFs exported from notation programs are ideal. Very small staff sizes, ornate engraving from old editions, and dense orchestral reductions are harder to read.
If a book has cover pages or long text sections, you can leave them in — pages without staves are skipped — but removing them makes conversion faster.
Typical errors
Common mistakes are wrong rhythms in a few bars (which make the bar 'overfull'), missed accidentals, ties read as slurs, grace notes misplaced, and voices merged incorrectly. Triplets and tuplets are a frequent source of rhythm errors.
After converting, play the piece slowly and watch for bars that sound wrong or drift out of time. Piano: Soloist points out bars whose rhythm adds up to more beats than the time signature allows, which is a quick way to find recognition errors.
Copyright and sharing
Converting scores you own for personal practice is common, but scanned editions of copyrighted music should not be shared. Many public-domain scores are available as free digital editions, which also convert much more accurately than scans.