Towards a model of loudness recalibration

Abstract

The Zwicker (1977, 1990) loudness model is a standard for predicting the loudness of a sound. This model, along with Moore and Glasberg's (see Acustica, vol.82, p.335-45, 1996) revision of it, is fairly accurate at predicting the loudness of steady-state sounds, but falls short for many types of temporally varying sounds. One temporal effect not accounted for in the Zwicker model is loudness recalibration. Loudness recalibration is a fatigue-like effect that makes a quiet tone at one frequency even quieter when it is preceded by a louder tone at the same frequency. The evidence suggests that loudness recalibration occurs in the central nervous system. Two means of modeling loudness recalibration are proposed. The first is an algorithmic description of the recalibration effect that could be added to the later stages of the Zwicker model. The other method uses a neural network and is based on a spike-train timing theory of hearing rather than a rate-place theory as assumed by the Zwicker model. This spike-train timing approach is unique in that spike-train averaging is postponed until a final loudness estimate is made. A more complete and accurate model of loudness recalibration will have to wait until more experimental data is collected.