Music Synthesis with Reconstructive Phrase Modeling

Abstract

This article describes a new synthesis technology called reconstructive phrase modeling (RPM). A goal of RPM is to combine the realistic sound quality of sampling with the performance interaction of functional synthesis. Great importance is placed on capturing the dynamics of note transitions-slurs, legato, bow changes, etc. Expressive results are achieved with conventional keyboard controllers. Mastery of special performance techniques is not needed. RPM is an analysis-synthesis system that is related to two important trends in computer music research. The first is a form of additive synthesis in which sounds are represented as a sum of time-varying harmonics plus noise elements. RPM creates expressive performances by searching a database of idiomatic instrumental phrases and combining modified fragments of these phrases to form a new expressive performance. This approach is related to another research trend called concatenative synthesis