Photograph or scan a printed score and get it back as data — editable MusicXML, playable MIDI, and a rendered view that highlights the notes as it plays.
OMR is OCR's harder cousin. Text runs left to right in one line; music is two-dimensional — pitch is vertical position, duration is notehead shape and stem, and a single symbol's meaning depends on the clef several centimetres to its left and a key signature that may have changed two systems ago.
The pipeline finds staff lines, removes them without destroying the symbols sitting on them, classifies what is left, then reconstructs the musical logic. We run Audiveris — the most capable open-source OMR engine — and add a rescue ladder: if a page comes back empty or short we retry it — auto-detecting and correcting rotation, flattening uneven lighting, straightening bent staff lines, splitting two-up spreads, upscaling small images and re-rasterising PDFs at 300 dpi — which recovers a lot of phone photos that would otherwise fail outright.
Recognition quality is decided before you upload. In rough order of impact:
Once it is digital you can transpose it, extract a single part, slow it down for practice, or edit it in MuseScore. Want just the notation file? PDF to MusicXML. Just the playable file? PDF to MIDI.