Posts Tagged ‘OCR’

Optical Music Recognition

Some time ago, the What’s the score at the Bodleian? project team went to see Matt McGrattan at The Bodleian Digital Library Systems and Services. We wanted to find out what it would take to be able to use our digitized scores to automatically generate a sound file to go with the sheet music, and Matt had been looking into this.

A number of programs exist that will convert images of music into a kind of notation that can be read by computers to, for example, generate a sound file or be used as input into music editing programs. Background reading on the matter had suggested that it was unlikely that our material would convert easily as far too many variables were non-ideal (some references in the Optical Music Recognition Bibliography). We nevertheless wanted to explore what it would take to make it worth-while to include automatic music recognition in the project.

Screenshot of Audiveris interface (from

Screenshot of Audiveris interface (from

The program Matt used for our initial test was Audiveris. Audiveris is an open-source Optical Music Recognition (OMR) tool that can ‘interpret’ music notation and convert it to a form of data (Music XML) that can then be used as input into other programs.

Before we could use the program, our sample file had to be pre-processed (for example making sure it was the right format and size). The file was then loaded into Audiveris and processed as illustrated in the Quick example found on the Audiveris website.

The initial output that we got was not perfect, and what this meant was obvious when the file was used to automatically generate a sound file. Matt suggested it sounded ‘like something by Scott Joplin’. For some kinds of music that is the desired effect, but in this case it was not. It is perfectly possible to post-edit the initial output and manually correct some of the problems, but the time and effort necessary for this means we could not fit it in to the current phase of the project. It is, however, something we want to continue to look into.

This test only included one program (Audiveris) and was performed on only one of our samples. It is possible that other programs will suit our material better, or that this process will be better suited for other types of material. As we are hoping to be able to digitize and make available other kinds of scores in the future, we will continue to explore options for automatic optical music recognition. We’ll report on any further findings as and when we have some.

%d bloggers like this: