EXPLAINING MUSICAL EXPRESSION AS A MIXTURE OF BASIS FUNCTIONS

The quest for understanding how pianists interpret notated music to turn it into a lively musical experience, has led to numerous models of musical expression. One of the major dimensions of musical expression is loudness. Several models exist that explain loudness variations over the course of a performance, in terms of for example phrase structure, or musical accent. Often however, especially in piano music from the romantic period, performance directives are written explicitly in the score to guide performers. It is to be expected that such directives can explain a large part of the loudness variations. In this paper, we present a method to model the influence of notated loudness directives on loudness in piano performances, based on least squares fitting of a set of basis functions. We demonstrate that the linear basis model approach is general enough to allow for incorporating arbitrary musical features. In particular, we show that by including notated pitch in addition to loudness directives, the model also accounts for loudness effects in relation to voice-leading.