On the way to ambient media for sheet music by techniques of information retrieval

Abstract

When the smart computing devices are more pervasive, the ambient media are more penetrating into daily lives. In order to make sheet music as ambient media, the research takes the important step to transform sheet music, Chinese hymn, into computer readable and playable form by optical music recognition (OMR) of information retrieval (IR) techniques. As one of music notation system, the numbered music notation is popular in many Asia countries. The building blocks and challenges to build an OMR system for this notation are illustrated in the study. The research proposed the music notation intermediate format (MNIF) for the OMR system for algorithm development and database construction. The study also devised the algorithms to improve the recognition rates of digit notation, pitch notation and length notation, including self-training learning model. Moreover, the study provided the metrics to evaluate the performance of OMR system for different stages and the longest common notes (LCN) based procedures to enable the scoring. The results of the evaluation on the proprietary OMR dataset show the performance level of the current note recognizer. Especially, to the best knowledge of the authors, the OMR dataset and the system are the first systematic construction to address the application.