A Machine Learning Approach for Analyzing Musical Expressions of Piano Performance

This paper proposed a machine learning approach for analyzing teacherspsila expert knowledge of classifying studentspsila piano performance into approximate expression categories. Students are usually confused when learning the expressive performance because of teacherspsila subjective intention difference on the same performance. In this paper, teacher models was built by analyzing teacherspsila classification rules. By replaying their performances and read teacherspsila suggestions in graphical and textual modes which are generated automatically by teacher model, students could understand the nuance of performance features on each expression. Three teachers and ten students joined this experiment. Sixty piano performances were recorded for constructing the teacher models. The average accuracy of teacher models for classifying performance expression is 70.8%. Questionnaires reflect both teachers and students are satisfied with the user interface, generated suggestions, and classification rules.