Grouping compositions based on similarity of music themes

PLoS One. 2020 Oct 8;15(10):e0240443. doi: 10.1371/journal.pone.0240443. eCollection 2020.

Abstract

Finding music pieces whose similarity is explainable in plain musical terms can be of considerable value in many applications. We propose a composition grouping method based on musicological approach. The underlying idea is to compare music notation to natural language. In music notation, a musical theme corresponds to a word. The more similar motives we find in two musical pieces, the higher is their overall similarity score. We develop the definition of a motive as well as the way to compare motives and whole compositions. To verify our framework we conduct a number of grouping and classification experiments for typical musical corpora. They include works by classical composers and examples of folk music. Obtained results are encouraging; the method is able to find non-obvious similarities, yet its operation remains explicable on the ground of music history. The proposed approach can be used in music recommendation and anti-plagiarism systems. Due to the musicological flavor, one of potentially best applications of our method would be that in computer assisted music analysis tools.

MeSH terms

  • Cluster Analysis
  • Humans
  • Sprache
  • Music*
  • Natural Language Processing*

Grants and funding

The authors received no specific funding for this work.