A Machine Learning Approach to Study Expressive Performance Deviations in Classical Guitar

Abstract

Expression is the added value of a musical performance, in which deviations in timing, energy, and articulation are introduced by musicians. Computational models have been proposed aiming at understanding and modelling the expressive content of music performances, to convey concrete expressive intentions. However, little work has been done to investigate the intrinsic variations that musicians might introduce, i.e. when no specific expressive indications are provided. In this contribution, we present a machine learning approach to study the expressive variations that nine different guitarists introduce when performing the same musical piece, for which no performance indications are provided. We study the correlations on the variations in timing and energy. We extract features from the score to obtain predictive models for each musician to later cross-validate among them. Preliminary results indicate that musicians use similar variations when applying these variations, based on correlation measures. Also, similar correlation indexes are found on the cross-validation exercise.