Most commercial Optical Music Recognition (OMR) systems work in a fairly opaque, inflexible way: the user puts an image in; it gives a file out. Academic OMR systems tend to be much more complex and are rarely used outside of a research lab.
This is where Rodan comes in—as a middle ground. Rodan is a web-based OMR system for users who want a more complex and adaptable system. The user puts an image in, but then can build a custom workflow for each processing step, creating a flexible approach to OMR systems. Once the preferred adjustments are customized for one image, the same processing can be applied to all images. Rodan’s customizable OMR process is designed to work with libraries, allowing them to perform OMR on millions of books.
Rodan uses a technique known as ‘machine learning’—teaching the machine to self-correct and to ‘learn’ from its mistakes. For music recognition purposes, machine learning is even more crucial, since there is any number of ways that a note can be printed or drawn.
Since libraries will require automatically recognized books with as few errors as possible, Rodan facilitates the collection of ‘ground truth data’ (human-corrected data) that is then fed back into its system. The corrected data is shared—so that individuals can collaborate on OMR project from around the world.
With text, we can search easily using initiatives like Google Book Search. Currently no such technology exists for searching music scores. Rodan is helping libraries build search systems for retrieving digital images of their collections through a “full music” (like “full text”) search. When a library’s million books are digitized and processed using Rodan, they can be mined for this extra-musical information, and researchers will be able to find, say, all the similar pieces to the one searched for.
Rodan technology is part of the larger SIMSSA (Single Interface for Music Score Searching and Analysis) project, under Principal Investigator Dr. Ichiro Fujinaga. Rodan is currently SSHRC funded with a Partnership Development grant.
Individuals interested in the Rodan project may contact Andrew Hankinson (email@example.com).