Abstract
Analysing expressive timing in performed music can help machine to perform various perceptual tasks such as identifying performers and understand music structures in classical music. A hierarchical structure is commonly used for expressive timing analysis. This paper provides a statistical demonstration to support the use of hierarchical structure in expressive timing analysis by presenting two groups of model selection tests. The first model selection test uses expressive timing to determine the location of music structure boundaries. The second model selection test is matching a piece of performance with the same performer playing another given piece. Comparing the results of model selection tests, the preferred hierarchical structures in these two model selection tests are not the same. While determining music structure boundaries demands a hierarchical structure with more levels in the expressive timing analysis, a hierarchical structure with less levels helps identifying the dedicated performer in most cases.