Enhanced multi-channel model for auditory spectrotemporal integration

In psychoacoustics, a multi-channel model has traditionally been used to describe detection improvement for multicomponent signals. This model commonly postulates that energy or information within either the frequency or time domain is transformed into a probabilistic decision variable across the auditory channels, and that their weighted linear summation determines optimum detection performance when compared to a critical value such as a decision criterion. In this study, representative integration-based channel models, specifically focused on signal-processing properties of the auditory periphery are reviewed (e.g., Durlach's channel model). In addition, major limitations of the previous channel models are described when applied to spectral, temporal, and spectrotemporal integration performance by human listeners. Here, integration refers to detection threshold improvements as the number of brief tone bursts in a signal is increased. Previous versions of the multi-channel model underestimate listener performance in these experiments. Further, they are unable to apply a single processing unit to signals which vary simultaneously in time and frequency. Improvements to the previous channel models are proposed by considering more realistic conditions such as correlated signal responses in the auditory channels, nonlinear properties in system performance, and a peripheral processing unit operating in both time and frequency domains.