Lessons Learned From a Project to Encode Mensural Music on a Large Scale With Optical Music Recognition

This paper discusses the transcription of a collection of musical works using Optical Music Recognition (OMR) technologies during the implementation of the Spanish PolifonIA project. The project employs a research-oriented OMR application that leverages modern Artificial Intelligence (AI) technology to encode musical works from images into structured formats. The paper outlines the transcription workflow in several phases: selection, preparation, action, and resolution, emphasizing the efficiency of using AI to reduce manual transcription efforts. The tool facilitated various tasks such as document analysis, management of parts, and automatic content recognition, although manual corrections were still indispensable for ensuring accuracy, especially for complex musical notations and layouts. Our study also highlights the iterative process of model training and corrections that gradually improved transcription speed and accuracy. Furthermore, the paper delves into challenges like managing non-musical elements and the limitations of current OMR technologies with early musical notations. Our findings suggest that while automated tools significantly accelerate the transcription process, they require continuous refinement and human oversight to handle diverse and complex musical documents effectively.