Reveal Hidden Information in the Music Scores: Composer Attribution

In this project, machine learning classi cation methods were applied to a music attribution challenge related to Josquin de Prez, a famous composer of the Renaissance period with many works misattributed to him. To solve the attribution problem, we trained classi ers using two kinds of music scores: scores veri ed to be composed by Josquin and scores composed by other contemporary musicians. These scores were retrieved from the Stanford Josquin Research Project. Features were extracted from the scores with knowledge of music theory and techniques of text mining. The optimized classi ers accurately distinguish whether a music score is composed by Josquin with a precision of 95% and a recall of 90%. An unsecure set of music works was analyzed by the optimized classi ers. From the successful results, our feature extraction and classi cation scheme may nd useful applications in more general composer attribution problems.