A NEW DISSIMILARITY METRIC FOR THE CLUSTERING OF PARTIALS USING THE COMMON VARIATION CUE
Mathieu Lagrange · 2005
In order to be able to operate relevant musical transformations of a sound, a knowledge of the musical structure of this sound is often mandatory. Therefore, we consider a structured representation of sound composed of acoustical entities on top of the widely used sinusoidal model. From the analysis point of view, these entities are clusters of partials which are perceived as a unique complex sound. To be able to automatically identify these clusters, numerous cues are proposed in the literature. We study in this article a generic one, the common variation of the parameters of the partials. Thanks to an original dissimilarity metric based on the autoregressive modeling of the vectors of frequency or amplitude of the partials, we are able to use even micro-modulations of these parameters for our purpose.