Modelling expressive performance using consistent evolutionary regression trees

TitleModelling expressive performance using consistent evolutionary regression trees
Publication TypeConference Paper
Year of Publication2006
AuthorsHazan, A., & Ramirez R.
AbstractWe present an evolutionary approach for building regression tree based models in the context of expressive music performance. We first review the benefits of using evolutionary computation techniques in this context. We then use the strongly-typed genetic programming framework and define the types and constraints that are needed for evolving efficiently multi-dimensional regression trees, and present two fitness functions for modelling expressive performance local timing. While the first fitness measurement is purely error-driven, the second also takes into account the balance of the evolved tree in terms of input space representation. Finally, we present the results of both learning and generalization experiments. For these experiments, we use a database of saxophone performance timing extracted from a set of acoustic recordings of jazz standards. The whole system is built into the Open Beagle evolutionary computation framework.
preprint/postprint documentfiles/publications/bec0a7-ecai2006-hazan.pdf