Optical Music Recognition with Rodan

This site describes the Optical Music Recognition (OMR) process implemented by the Distributed Digital Music Archives and Libraries (DDMAL) lab for encoding manuscripts in the Music Encoding Initiative format. It demonstrates the Rodan workflow builder and manager and stages of processing for interpreting and encoding square-notation music with machine learning.

General information about OMR is available on the main overview page, which also includes subpages on the different steps of the OMR process. Each subpage goes into detail on specific jobs that accomplish a specific task. Additionally, there is a general overview of Rodan.

New users can follow a brief tutorial of performing OMR using the CDN-Mlr 073 manuscript as an example, which is available in the tutorial pages.

What Do I Need To Get Started?

Hardware and Software

Rodan is a web app and therefore doesn’t require you to install anything, but for the best experience, it’s recommended that you use a recent version of Google Chrome or Firefox and use a computer with at least 8 GB of RAM. However, everything should work on a modern computer using a recent version of a popular browser.

It may be necessary to generate some resources locally (e.g., like the Optical Character Recognition models.) Instructions to create these files are available with the jobs documentation.

Digital Resources

This largely depends on your goals, but if you intend to create an encoding of a page or a few pages in the Music Encoding Initiative (MEI) format, you’ll need the following materials:

  1. High-quality images of the manuscript pages;
  2. Computational models trained to detect score elements in this kind of manuscript: staff lines, text, and music symbols.
  3. Training data for classifying music symbol glyphs in the format used by Gamera;
  4. A CSV file mapping classes of glyphs to fragments of MEI; and
  5. A plain-text transcript of the neumed text or lyrics on each page of the manuscript and an OCR model capable of reading that text.