Main Article Content
Musical rhythm is a complex experiences, which is structured in time. Furthermore, every musical event has a distinct sound. Thus, it is plausible that investigations in rhythm have to consider its sound, too. Commonly, sounds are discriminated by their timbre. Therefore, rhythm can be described as succession of distinct timbres. We developed a method to model drum patterns in such a manner. Timbre is approximated as a one-dimensional feature consisting of weighted spectral centroid. An onset detection algorithm based on fractal geometry determines the time frames of measurement within the input audio file. The resulting time series is used to train an m-state Hidden Markov Model. The model’s transition probability matrix serves as a fingerprint of the sample’s rhythm. This method can therefore be used to compare music quantitatively and to reveal and cluster musical similarities in sound recording archives.