Human information processing and auditory scene analysis

The effects of extraneous stimulus components (interferers) on listeners' sensitivity to changes in a target component are measured in a series of discrimination experiments and discussed in terms of quantitative models for discrimination. Specific attention is given to comparisons between conditions in which that target component is perceptually fused with the interferers and conditions in which the target component is perceived as a separate auditory object. Basic notions of auditory scene analysis are tested in these quantitative comparisons. Listeners' abilities to discriminate changes in the intensity or interaural time delay (ITD) of a target sinusoid were measured in several conditions. These conditions included the target sinusoid presented in isolation, presented with components that perceptually fuse with the target, and presented with components that result in separately perceived target and interferer auditory objects. The interferers were either sinusoidal or noise stimuli, and performance was measured at 70 and 40 dB SPL, with 400- and 200-Hz sinusoidal interferer spacings, and with continuous and pulsed interferers. Just noticeable differences (jnds) increased when spacing narrowed, and were greater in one interferer condition (sinusoidal, continuous, wide-spacing, low-level) were the jnds the same as those of the target-alone condition for each subject. A new model for ITD processing was developed to understand this interference. The model uses specification of the input component amplitudes, frequencies, and ITD values, and a peripheral auditory filtering specification, to produce a frequency-dependent representation of ITD based on long-term cross-correlation. A linear combination across frequency, dependent upon the perceptual parsing of stimulus components, is shown to describe successfully ITD discrimination data from the literature, and the ITD discrimination results with tonal interferers in this study. It is found that the model uses channels over a range of frequencies much wider than expected (i.e., under than one-third octave) in processing a tone. Increased jnds in the model results for the interferer conditions are mainly due to decreases in the delay at which the peak of the correlation function occurs in a portion of the frequency channels.