SIGNAL DETECTION ENHANCED BY COMODULATED NOISE

Signal detection in fluctuating background noise is a common problem in diverse fields of research and technology. It has been shown in hearing research that the detection of signals in noise that is correlated in amplitude across the frequency spectrum (comodulated) can be improved compared to uncorrelated background noise. We show that the mechanism leading to this effect is a general phenomenon which may be utilized in other areas where signal detection in comodulated noise needs to be done with a limited frequency resolution. Our model is based on neurophysiological experiments. The proposed signal detection scheme evaluates a fluctuating envelope, the statistics of which depend on the correlation structure across the spectrum of the noise. In our model, signal detection does not require a sophisticated neuronal network but can be accomplished through the encoding of the compressed stimulus envelope in the firing rate of neurons in the auditory system.

Influence of Frequency Selectivity on Comodulation Masking Release in Normal-Hearing Listeners

Experiments 1 and 2 investigated the effect of frequency selectivity on comodulation masking release (CMR) in normal-hearing subjects, examining conditions where frequency selectivity was relatively good (low masker level at both low [500-Hz] and high [2500-Hz] signal frequency, and high masker level at low signal frequency) and where frequency selectivity was somewhat degraded (high masker level and high signal frequency). The first experiment investigated CMR in conditions where a narrow modulated noise band was centered on the signal frequency, and a wider comodulated noise band was located below the band centered on the signal frequency. Signal frequencies were 500 and 2000 Hz. The masker level and the frequency separation between the on-signal and comodulated flanking band were varied. In addition to conditions where the flanking band and on-signal band were presented at the same spectrum level, conditions were included where the spectrum level of the flanking band was 10-dB higher than that of the on-signal band, in order to accentuate effects of reduced frequency selectivity. Results indicated that CMR was reduced at the 2000-Hz region when masker level was high, when the frequency separation between on-signal and flanking band was small, and when a 10-dB level disparity existed between the on-signal and flanking band. In the second experiment, CMR was investigated for narrow comodulated noise bands, presented either without any additional sound or in the presence of a random noise background. CMR increased slightly as the masker level increased, except at 2500 Hz when the noise background was present. The decrease in CMR at 2500 Hz with the high masker level and with a noise background present could be explained in terms of reduced frequency selectivity. In a third experiment, we compared performance for equal absolute bandwidth maskers at a low (500-Hz) and a high (2000-Hz) stimulus frequency. Results here suggested that detection in modulated noise may be reduced due to a reduction in the number of quasi-independent auditory filters contributing temporal envelope information. The effects found in the present study using normal-hearing listeners under conditions of degraded frequency selectivity may be useful in understanding part of the reduction of CMR that occurs in cochlear-impaired listeners having reduced frequency selectivity.