A music similarity measure based on chord progression and song segmentation analysis

Music documents retrieval system in most websites are useful for users who want to search sheet music which consists of chords, notes, or lyrics in a printable sheet format. With the growth of music repositories, music classification based on musical similarity can help users to find the same kind of songs. One important aspect to define music similarity measure is chord progression. In this paper, we present an approach to extract chord progressions from song segmentation in chord sheet music documents. This approach uses signature files to encode chords information with preserved musical data. Then, we propose a novel model for chord sheet music similarity without beats information which composes of three computation levels: (1) chord similarity, (2) sequence similarity, and (3) music similarity. In addition, we improve efficiency by removing duplicated parts. For the evaluation, we set an experiment to compare our music similarity model with others. The results indicated that music similarity based on sequences of chords obtained higher precision than chord classification based on the frequencies of chords. Furthermore, our model can sustain effectiveness even if beats information are unknown. We found that segmentation of chord progression can be another important aspect of musical similarity.