Modeling the detection and level discrimination of tones in the presence of masking noise

Abstract

The performance of three model detectors was evaluated for masked detection and level discrimination of tones in noise with different bandwidths and in a roving level paradigm (Kidd et al., 1989). The model detectors were based on (1) the output energy of several auditory filters tuned to different frequencies, (2) the envelope statistics (peakiness) of the auditory filter outputs, and (3) the cross-correlations between auditory-nerve (AN) fibers in a population model. The energy-based model predicted detection thresholds, but failed to predict masked level discrimination; this detector was robust for roving-levels in the wideband condition by combining information from different auditory filter outputs and the prediction was affected by roving narrow-band masker levels. Thresholds of the envelope-based detector were worse than human thresholds for some conditions; however, this model was robust in roving-levels for both wideband and narrow-band maskers. The monaural cross-correlation detector included model cells that were sensitive to temporal cues in model AN inputs in response to wideband noise and also included cells that were sensitive to input level changes. This model naturally combined rate and temporal information and predicted performance for both masked detection and discrimination. [Work supported by NIDCD R01-01641.]