For my Masters dissertation (which will be built upon for my thesis) I aim to look at how music recommendation systems can be enhanced by using musical analysis (especially harmonic analysis) of encoded data. Optical Music Recognition (OMR) is one way of providing this encoded data, and thus I began my research with looking at OMR. Here is a summary of some of my research into OMR. And what I have learnt, and built upon from it.
What is optical music recognition
Doing OMR is hard. Very hard. (Gerd Castan)
OMR is a process of converting scanned pages of music into an online format, enabling computers to ‘read’ and manipulate printed music. OMR is an extension of Optical Character Recognition (OCR), which looks similarly at converting scanned pages of text into an on-line format. OCR is better established than that of OMR – which is still in development. OCR has enabled textual documents to be searched, as used by Google Books – where you can search for a specific word within a document. OMR therefore could enable Music to become searchable, by creating encoded musical data. Encoded musical data could enable musical analysis to be completed online, using computational programs to compare the harmony, texture and rhythmic components of pieces of music. By enabling this, musical data could be added to the factors of recommendation, instead of purely relying on metadata- the current process. OMR is not the only way of creating this encoded data, the mark-up languages such as the Music Encoding Initiative, is another such way of doing this.
A brief history of OMR research
Early OMR research, looked at creating systems which could recognise Common Music Notation (CMN). Pruslin (1966), Kassler (1972), Prerau (1970, 1971, 1975) all looked at features of CMN such as clefs, time signatures and rests. Prerau built upon the work of Pruslin and Kassler in his later work, and established a system that recognised clefs, rests, certain time signatures and accidentals. A large amount of research has also looked into the music recognition of hand written manuscripts. However OMR is yet to reach the success of OCR through Google Books and HathiTrust. SIMSSA (Single Interface for Music Score Searching and Analysis) is a current project, that aims to develop OMR’s capabilities. The Cantus Ultimus project (part of SIMSSA) is an example of their OMR work in action, using chant manuscripts. This project is one such area of music which OMR will be used on, when successfully developed. The Cantus Ultimus project shows you, how OMR will enable searching and the success that the SIMSSA project is having.
How does this relate to my dissertation?
OMR is one way which individuals can change sheet music into encoded data enabling musical analysis and searchability of music. OMR could enable recommendation systems to further adapt their algorithms to look at the musical data present. OMR is one way of enabling musical comparisons to be created, so that we can use computers to establish how music is similar or different to each other. These similarities and differences could include looking at harmonic progressions, rhythmic patters, and textural aspects.
Bullen, A. 2008. Bringing Sheet Music to Life: My Experiences with OMR. Code4Lib, 3, [online] Available at: <http://journal.code4lib.org/articles/84> [Accessed: 22.01.2016].
Bainbridge, D. and Bell, T. 2001. The Challenge of Optical Music Recognition. Computers and Humanities, 35, pp. 95-121.
Castan, G. 2012. Optical Music Recognition (OMR). Music-Notation, [online] Available at: <http://www.music-notation.info/en/compmus/omr.html> [Accessed: 22.01.206].
SIMSSA, [online] Available at: <https://simssa.ca/about> [Accessed 25.01.2016].