Evaluating OMR on the Early Music Online Collection

The Early Music Online (EMO) collection consists of about 300 printed music books of the sixteenth century held at the British Library. They were recently digitized from microfilms and made available online. In total, about 35,000 pages were digitized. This paper presents an optical music recognition (OMR) evaluation on the EMO collection. Firstly, the content of the collection is reviewed, looking at the type of music notation and the type of printing technique. Secondly, for the books for which it is possible (260 books), an OMR evaluation performed using the Aruspix OMR software application is presented. For each book, one randomly selected page of music was processed and the recognition rate was computed using a corrected transcription of the page. This evaluation shows very promising results for large-scale OMR on the EMO or similar collections. The paper also highlights critical points that should be taken into account in such an enterprise.