Contour representations of sound

We consider how the continuity of form in natural sounds may be used to discover sparse time-frequency representations. To proceed, we describe a method to represent any sound as a collection of contours in the time-frequency plane. By analysing the signal in many time-scales, an over-complete set of shapes is generated for a given sound. From this redundant set of shapes the simplest, most parsimonious mathematical forms may be selected.