Modelling Gestures in Music Performance with Statistical Latent-State Models

"We discuss how to model ""gestures"" in music performance with statisticallatent-states models. A music performance can be described with compositionaland expressive properties varying over time. In those property changes we oftenobserve particular patterns, and such a pattern can be understood as a""gesture"", since it serves as a medium transferring specific emotions. Assuminga finite number of latent states on each property value changes, we candescribe those gestures with statistical latent-states models, and train themby unsupervised learning algorithms. In addition, model entropy provides us ameasure for different effects of each properties on the gesture implementation.Test result on some of real performances indicates that the trained modelscould capture the structure of gestures observed in the given performances, anddetect their boundaries. The entropy-based measure was informative tounderstand the effectiveness of each property on the gesture implementation.Test result on large corpora indicates that our model has potentials for afurther model improvement."