Pitch range: C0-C9
Supported note pitches span the full practical range from C0 through C9, subject to the active clef and written accidentals.
SyMusOMRcodename turns score images into editable musical notation using a compact, on-device MusicOMR engine. Explore what the current beta recognizes, review known boundaries, read about the research, and request an invitation to test it.
The beta is under active development. Recognition coverage and behavior may change as testing expands.
The engine reads musical relationships across the page before constructing an editable score. Dense passages can take longer because more elements must be related and decoded.
Supported note pitches span the full practical range from C0 through C9, subject to the active clef and written accidentals.
Single and multiple beams are recognized through 32nd-note values when stems and beams are visually connected.
Tuplet groupings and single- or multi-dotted durations are interpreted as musical timing, not merely as separate marks.
Independent upper and lower voices can be imported from the same staff, including opposing stem directions.
The beta is designed for both exported/scanned pages and camera images. Perspective, rotation, lighting, and noise can be tolerated, but cleaner inputs remain faster and more reliable.
Handles clean exports and scans with common scanner noise, compression, and moderate skew.
Supports photos with moderate perspective distortion, rotation, and uneven lighting.
Current beta target: images from roughly 100 x 100 through 2000 x 2000 pixels. Avoid extreme downsampling of dense scores.
Filled or open oval noteheads, connected stems and flags, conventional beam thickness, and clearly marked rests work best.
The engine can normalize common viewing angles; severe keystone distortion or cropped systems may reduce accuracy.
Even illumination, readable staff lines, and adequate contrast reduce ambiguity and processing time.
SyMusOMR first builds an understanding of the page, then writes the score. Processing time generally grows with the number of staves, voices, notes, rests, and structural changes - not just the image dimensions.
Some notations carry both musical meaning and page-layout meaning. In the present beta, the engine prioritizes the playable/editable musical result.
Notes, rests, pitches, durations, chords, and beam content are recognized. On import, a note may remain assigned to its originating staff rather than reproducing the visual crossing exactly.
Within one page, one system may contain a different number of staves than the next. SyMusOMR does not assume a fixed staff count across the page: it infers each system from the image as it reads. Dense layouts have been tested reliably, though note accuracy can fall below the ~90% seen on clean, larger, clearly written music when systems are unusually tight, small, or cropped.
These boundaries help prospective testers understand which results are currently expected and which capabilities remain on the roadmap.
Staff position and duration are imported, but the specific x, circle-x, triangle, slash, parenthesized, or other custom notehead symbol is not yet preserved.
Textual chord symbols and sung lyrics are not part of the current beta recognition set.
Guitar and other tablature systems are planned but are not imported by this beta.
Titles, composer names, part names, and free-form page annotations are not presently recognized as score metadata.
Dynamics, articulations, expression text, and other special marks not listed as supported should be treated as roadmap items.
Beta feedback will help prioritize commonly encountered symbols, score types, and layouts.
Capture the full system; keep staff lines sharp; avoid motion blur and hard shadows; use the highest practical resolution; and leave a modest margin around the music. For handwriting, connect stems, flags, and noteheads clearly.
Beta behavior may change as recognition coverage expands. Participants will receive guidance for distinguishing unexpected omissions from features that are still on the roadmap.
We are preparing SyMusOMR for broader testing across printed scores, camera images, scans, multi-staff music, and clean handwriting. Early testers can help identify the score types and notation features that matter most in real musical workflows.
Interested in the architecture and research direction? Read How the MusicOMR Engine Works.