Leaky integrators, Fletcher curves, and period histograms: A new link

Abstract

Fletcher curves, thresholds from tone-in-noise detection tasks plotted as a function of masker bandwidth, are thought to reflect characteristics of peripheral filters. This notion, however, is compromised by the finding that a roving level procedure affects thresholds only for very narrow bandwidths [Kidd et al., J. Acoust. Soc. Am. 86, 1310–1317 (1989)]. Alternatively, a three-stage, leaky integrator (i.e., initial, relatively broad bandpass filter, followed by half-wave rectification and low-pass filtering) yields frequency-dependent, level-invariant Fletcher curves, thus entertaining the possibility that sensitivity to envelope cues underlies detection at all masker bandwidths. The decision statistic is a measure of similarity (Mahalanobis distance) between the observed amplitude spectrum of the leaky integrator output and expected spectra with and without a signal. Regarding physiological plausibility, the model also describes the temporal response of auditory nerve fibers to complex stimuli. Horst et al. [J. Acoust. Soc. Am. 85, 1898–1901 (1985)] used multicomponent, periodic stimuli to obtain period histograms from auditory nerve recordings. Fourier transforms of the histograms were used to assess temporal responses of fibers. For various stimulus configurations, the spectra of the period histogram and leaky integrator output are nearly identical. Theoretical implications are discussed. [Work supported by ONR.]