Binaural models for the detection of interaural incoherence

Abstract

Human listeners have a remarkable ability to detect small amounts of interaural incoherence in bands of noise. Experiments using frozen noise, all with the same value of interaural coherence (0.9922), show that when the bandwidth is large the ability to detect incoherence is essentially determined by the value of coherence itself. However, when the bandwidth is narrow, e.g. 20 Hz, detection depends on some function of the moment-to-moment fluctuations in interaural phase differences and interaural level differences. Experiments using hundreds of different frozen noises and two bandwidths were performed to find the best binaural model for predicting detection performance for such narrow bands. Successful models incorporated known elements of binaural hearing—temporal averaging and compression. Models with independent processing of interaural phase fluctuations and interaural level fluctuations were marginally more successful than models based on fluctuations in lateral position. The most successful models correlated well with average human responses, r=0.89. It seems likely that such individual fluctuations remain dominant in masking level difference experiments in the narrow band limit. [Work supported by the NIDCD grant DC 00181.]