Expressive timing and dynamics are an important part of musical meaning. For those skeptical of this claim I will play some examples of music with and without expressive timing and dynamics. I will then provide an overview of previous approaches to measuring and modeling musical expression in different contexts. One relatively unexplored task is that of analyzing music for which no musical score is available. I will describe why this is a different problem than score following and will argue for the importance of score-free models in working with improvisation and related tasks. I will provide some preliminary results employing a correlation-based model (Autocorrelation Phase Matrix; APM) to infer metrical trees in unscored music. Expressive timing and dynamics can then be measured with respect to these trees.

A note for the audience:
The spirit of NIPS workshops is to bring new and perhaps even half-baked ideas to the table. In that spirit, my hope is that my talk is derailed into a discussion about measuring and modeling expressive timing and dynamics in unscored streams of MIDI or audio.