A machine learning approach to expressive performance in jazz standards

Abstract

We describe an approach to perform expressive transformation in monophonic Jazz melodies. The system consists of three components (a) a melodic transcription component which extracts a set of acoustic features from monophonic recordings, (b) a machine learning component which induce expressive transformation models from the set of extracted acoustic features, and (c) a melody synthesis component which generates expressive monophonic output (MIDI or audio) from inexpressive melody descriptions using the induced expressive transformation model. We describe and compare different machine learning methods for inducing the expressive transformation models.