A computational model of temporal integration

Abstract

In this paper a computational model of temporal integration is demonstrated. The model has the following architecture. The first stage computes a representation of the auditory nerve response based on the Sheffield ear. The nerve response is summed and low-pass filtered with a temporal window of about 10 ms [Plack and Moore, J. Acoust. Soc. Am. 87, 2178–2187 (1988)]. In the second stage the summed and smoothed nerve response is coupled to a transmission line model of auditory sensory memory. Close to the periphery the impulse response of the transmission line is sharp. As a pulse transmits into the system though, it becomes progressively more attenuated and spread out in time. In the third stage the line is tapped at different points (approximately 10-ms spacing) and peak responses detected, which then become input samples for higher order leaky integrator with a time constant of about 200 ms. The model thus resembles the ‘‘multiple looks’’ model as proposed by Viemeister [Viemeister and Wakefield, J. Acoust. Soc. Am. 90, 858–865 (1991)] and is able to account for the ‘‘resolution-integration’’ paradox. Quantitative predictions of the model are shown for the temporal integration example from the Houtsma, Rossing, and Wagenaars compact disk.