In this document, we present work on visual score analysis that has been performed in the context of the TROMPA project. We describe existing systems and our experience with them, as well as techniques that we have created and implemented as alternatives to existing optical music recognition (OMR) systems.
Sheet music plays an important role in Western music, and considering currently available online public-domain resources, many PDF documents with sheet music scans can be found. While musicians already extensively play from such scores, they also have important potential for increasing digital music accessibility and enrichment, e.g. by being synchronized to recordings, by becoming searchable for motives or patterns, or by being offered as a flexible digital edition allowing for annotations. Such applications all will require the extraction of musical information from the visual information in the scanned score. At the same time, they may not all require for a full OMR pipeline to be run (i.e., going from a full PDF to a full transcription). Furthermore, considering TROMPA’s interest in human-in-the-loop approaches, rather than running full OMR pipelines which will always need post-correction, other hybrid annotation workflows (as also discussed in D4.4) are possible in which intermediate output of an OMR pipeline can already be transcribed, corrected or annotated. This deliverable also focuses on extracting such intermediate outputs from a PDF file, which is the most common container format for digital sheet music.