PAD and SAD: Two awareness-weighted rhythmic similarity distances

Abstract

Measuring rhythm similarity is relevant for the analysis and generation of music. Existing similarity metrics tend to consider our perception of rhythms as being in time without discriminating the importance of some regions over others. In a previously reported experiment we observed that measures of similarity may differ given the presence or absence of a pulse inducing sound and the importance of those measures is not constant along the pattern. These results are now reinterpreted by refining the previously proposed metrics. We consider that the perceptual contribution of each beat to the measured similarity is non-homogeneous but might indeed depend on the temporal positions of the beat along the bar. We show that with these improvements, the correlation between the previously evaluated experimental similarity and predictions based on our metrics increases substantially. We conclude by discussing a possible new methodology for evaluating rhythmic similarity between audio loops.