Abstract
Phrases are common musical units akin to that in speech and text. In music performance, performers often change the way they vary the tempo from one phrase to the next in order to choreograph patterns of repetition and contrast. This activity is commonly referred to as expressive music performance. Despite its importance, expressive performance is still poorly understood. No formal models exist that would explain, or at least quantify and characterise, aspects of commonalities and differences in performance style. In this paper we present such a model for tempo variation between phrases in a performance. We demonstrate that the model provides a good fit with a performance database of 25 pieces and that perceptually important information is not lost through the modelling process.