Optical Music Recognition on the International Music Score Library Project

FAIN: HD-51087-10

Trustees of Indiana University (Bloomington, IN 47405-7000)
Christopher Raphael (Project Director: March 2010 to February 2012)

Development of a prototype optical music recognition (OMR) software application and editorial platform to allow greater scholarly access to digitized music archives.

For decades scholars in computational musicology, music informatics, and other related disciplines have bemoaned the lack of symbolically-represented music. Such representations allow music to be searched, compared, transformed, and analyzed in myriad ways. Over the last several years the International Music Score Library Project (IMSLP), an open library of primarily scanned public domain classical music scores, has achieved "viral,'' status, now well-known in musical circles worldwide. This library allows easy and universal access to a wide and rapidly deepening collection of scanned musical scores. Optical music recognition (OMR), analogous to optical character recognition (OCR), converts music score images into symbolic form. We see in the IMSLP a potential gold mine of symbolic music data and propose, in conjunction with Indiana University's music library, to begin a long-term project toward developing the open source software needed to accomplish this goal.

Associated Products

New Approaches to Optical Music Recognition (Conference Paper/Presentation)
Title: New Approaches to Optical Music Recognition
Author: Christopher Raphael and Jingya Wang
Abstract: We present the beginnings of a new system for optical music recognition (OMR), aimed toward the score images of the International Music Score Library Project (IMSLP). Our system focuses on measures as the basic unit of recognition. We identify candidate composite symbols (chords and beamed groups) using grammatically-formulated top-down model-based methods, while employing template matching to find isolated rigid symbols. We reconcile these overlapping symbols by seeking non-overlapping variants of the composite symbols that best account for the pixel data. We present results on a representative score from the IMSLP.
Date: 10-25-11
Primary URL: http://www.music.informatics.indiana.edu/papers/ismir11/
Primary URL Description: Examples of the current state of the art for our recognition system.
Conference Name: International Symposium on Music Information Retrieval (ISMIR11)

---