Beyond the beat: modeling metric structure in music and performance.

Current models for capturing metric structure of recordings of music are concerned primarily with the task of tempo and beat estimation. Even though these models have the potential for extracting other metric and rhythmic information, this potential has not been realized. In this paper, a model for describing the general metric structure of audio signals and behavioral data is presented. This model employs reson filters, rather than the comb filters used in earlier models. The oscillatory nature of reson filters is investigated, as they may be better suited for extracting multiple metric levels in the onset patterns of acoustic signals. The model is tested with several types of sequences of Dirac impulses as inputs, in order to investigate the model's sensitivity to timing variations and accent structure. The model's responses to natural stimuli are illustrated, both for excerpts of recorded music from a large database utilized by tempo-estimation models, and sequences of taps from a bimanual tapping task. Finally, the relationship of the model to several other beat-finding and rhythm models is discussed, and several applications and extensions for the model are suggested.