Abstract
In computational musicology research, clustering is a common approach to the analysis of expression. Our research uses mathematical model selection criteria to evaluate the performance of clustered and non-clustered models applied to intra-phrase tempo variations in classical piano performances. By engaging different standardisation methods for the tempo variations and engaging different types of covariance matrices, multiple pieces of performances are used for evaluating the performance of candidate models. The results of tests suggest that the clustered models perform better than the non-clustered models and the original tempo data should be standardised by the mean of tempo within a phrase.