Ronald van Elburg - Structure in Sound: Texture Analysis

What constitutes structure in sound? Can we access sound structure in cochlear representations? We take a step back from current state of the art and formulate new answers to these questions.

In the analysis of sound generally the first or second processing step is a form of time-scale or time-frequency analysis, leading to a representation resembling that of a musical score. Part of the structure found in the resulting representation is due to the particular analysis used. We developed measures for structure in the time-scale representations which cannot be attributed to the time-scale analysis used and hence are attributable to structure in the source. This allows us to analyse the texture of sound, which we argue is woven from pulses, tones, rustles and reverberations. These representations give rise to relations between sounds which resemble the perceptual organization of sound in humans.