Desain, P. and Honing, H. (1997) How to evaluate generative models of expression in music performance. . In Isssues in Ai and Music Evaluation and Assessment. International Joint Conference on Artificial Intelligence., 5-7. Nagoya: Japan.

Abstract An outline is given of structural expression component theory (SECT). It generalizes existing generative models of music performance expression and extends them to deliver representations of musical timing and dynamics that can be composed into overall expression profiles. These compound profiles can than be fitted to actual performances. Based on this theory a technique is investigated to separate the expressive profiles of a musical performances into their structural components, each explained by one generative model. This method (DISSECT) allows us to obtain appropriate parameter setting for the individual models and effectively separates the expressive signal into its components, each explained by one kind of musical structure. Thus the explanatory power of each model can be evaluated, and the amounts of variation that can be attributed to different musical structural descriptions can be estimated for different musical styles and for various interpretations.

Full paper.