Music OCR

From Wikipedia, the free encyclopedia

Main article: Optical character recognition

Music OCR is the application of optical character recognition to interpret sheet music or printed scores into editable and, often, playable form. Once captured digitally, the music can be saved in commonly used file formats, e.g. MIDI (for playback) and MusicXML (for page layout).

Contents

[edit] History

Early research into recognition of printed sheet music was performed at the graduate level in the late 1960's at MIT and other institutions. (Pruslin, Dennis Howard (1966). "Automatic Recognition of Sheet Music" (PDF). Retrieved on 2007-01-24.) Successive efforts were made to localize and remove musical staff lines leaving symbols to be recognized and parsed. The first commercial music-scanning product, MIDISCAN, was released in 1991 by Musitek corporation.

Unlike OCR of text, where words are parsed sequentially, music notation involves parallel elements, as when several voices are present along with unattached performance symbols positioned nearby. Therefore, the spatial relationship between notes, expression marks, dynamics, articulations and other annotations is an important part of the expression of the music.

Modern music OCR packages have accuracy exceeding 99% when a clean scan is used and the notation is not exceptional (e.g. unfilled voices, non-standard symbology, etc.) Because music notation utilizes dots for staccato marks or to extend the value of a note, artifacts in the scan can lead to interpretation problems.

On June 7 2007, Neuratron Ltd (London, England) released the world's first commercial handwritten music OCR software, as integrated into its PhotoScore Ultimate 5 product which also became the world's first music scanning product to incorporate two printed music recognition engines, correlating the results of each to achieve significantly improved accuracy.

[edit] Proprietary software

[edit] Free/Open Source Software

[edit] See also