Modeling Perception of Spectral Profile Changes

In this thesis, we explore how the human auditory system represents and detects changes in a spectral profile. First, using profile analysis methods, we measure listeners' sensitivities to changes in spectral peak shapes and ripple phases. More specifically, we measure thresholds to changes in peak symmetry and bandwidth (which respectively are measures of the local evenness or oddness of a peak and of the tuning or sharpness of a peak). The effects of several other manipulations are also studied. It is found that the thresholds are constant for almost all initial peak shapes. Second, these changes in symmetry and bandwidth are interpreted as changes in the phase and magnitude of the profile's Fourier transform. In this light, the last set of experiments measured the sensitivity to (ripple) phase changes in spectral sinusoids. We find that the thresholds obtained are similar to the above-mentioned symmetry thresholds.A fundamental conclusion arising from this analysis is that spectral peaks are represented along two largely independent axes: the magnitude and phase of their Fourier transforms. More specifically, it is argued that, along these two dimensions, the auditory system analyzes an arbitrary spectral pattern in a localized Fourier transform domain. This is closely analogous to spatial frequency transformations in the visual system. Within this general framework, we propose a model of profile analysis in which a spectral profile is represented by a weighted sum of sinusoidally modulated spectra (ripples). The first part of the analysis is performed by a bank of bandpass filters, each tuned to a particular ripple frequency and ripple phase. The parameters of the model are estimated using data from several ripple discrimination experiments. The second part of the model is a detection stage which operates on the magnitude and phase of the computed transform, and varies with the type of perceptual task. The results of the detection operations are compared to experimental data from various profile analysis tasks. The model accounts well for the perceptual results in these tests. We propose two types of psychoacoustical experiments involving any arbitrary spectral patter, which should further verify the predictions of the model