SaxEx: a case-based reasoning system for generating expressive musical performances

Abstract

We have studied the problem of generating expressive musical performances in the context of tenor saxophone interpretations. We have done several recordings of a tenor sax playing dierent Jazz ballads with dierent degrees of expressiveness including an inexpressive interpretation of each ballad. These recordings are analyzed using SMS spectral modeling techniques to extract information related to several expressive parameters. This set of parameters and the scores constitute the set of cases (examples) of a case-based system. From this set of cases, the system infers a set of possible expressive transformations for a given new phrase applying similarity criteria, based on background musical knowledge, between this new phrase and the set of cases. Finally, SaxEx applies the inferred expressive transformations to the new phrase using the synthesis capabilities of SMS.