Stimulus-driven, time-varying weights for comodulation masking release.

This study tests the hypothesis that comodulation masking release (CMR) is mediated by "listening in the valleys" [S. Buus, J. Acoust. Soc. Am. 78, 1958-1965 (1985)]. Detectability was measured for signals consisting of six consecutive 25-ms, 1-kHz tone pulses presented in a 50-Hz-wide masker or in maskers consisting of seven 50-Hz-wide noises, one critical band apart, with either correlated or uncorrelated envelopes. The level of each signal pulse varied randomly around masked threshold according to Gaussian distributions with rms perturbations (standard deviations) of 3 or 6 dB. For each listener and condition, the responses from 5000 trials were sorted to construct conditional psychometric functions for d' as a function of signal-pulse intensity for ten ranges of short-term level of the on-frequency masker band during the pulse. The slopes of these functions for three normal listeners decrease markedly with increasing short-term masker level for the correlated multiband masker, but are largely constant for the other maskers. This indicates that the weight applied to the signal channel is high when the masker level is low and vice versa for the correlated masker, but is approximately constant for single-band and uncorrelated multiband maskers. These findings provide direct evidence that CMR is mediated by "listening in the valleys," but models based on direct envelope comparison may also account for the results if they are modified to include a compressive nonlinearity before the comparison.