Abstract:
At present, music search websites usually provide search function using music metadata such as titles, albums, artists, etc. In the case of searching a music document, sheet music can play an important role in a music retrieval system. The system will have more capability in searching music if the system can classify songs using chord sequences containing in sheet music documents. However, chord sequences extraction and songs similarity measurement seem to be difficult and challenging because of its uncertain text format. This research aims to propose a music similarity measure for sheet music documents considering three levels of calculation including chord, sequence, and music. In addition, our research proposes an approach of chord sequence extraction from the documents using text analysis based on music theory. Furthermore, we also propose a beat notation in the documents to specify beat information of chords. By the experimental results, our proposed method of music similarity measurement using chord progression obtained higher precision than using chord histogram technique about 53.97%. In addition, we found that an enhanced music similarity measurement considering beat notations added in the documents could improve the precision about 32.12%. For this research, we concluded that beat information and song segmentation are important factors for the precision of music similarity measure.