Interaural Correlation Discrimination with Spectrally-Remote Flanking Noise: Constraints for Models of Binaural Unmasking

Two experiments were used to discriminate between models of binaural unmasking. In each, interaural correlation sensitivity was measured for one-ERB-wide bands of noise centered at 500 Hz in the presence of remote flanking bands of correlated or anticorrelated noise. The flanking bands were separated from the target band on each side by spectral notches of 0, 1, 2, or 3 ERBs. Notches of 1 ERB or more produced smaller d′s for comparisons of widely differing positive and negative correlations (e.g. +0.8 and −1) than for smaller differences, particularly values compared to zero. These results are markedly at odds with most models of binaural processing. Experiment 1 used a 2I-2AFC loudness-discrimination task without feedback for 11 pairs of target-band correlation values. Fitted continuous cumulative d′ functions invariably showed peaks at a target-band correlation close to zero when notches were present; zero correlation sounded loudest. In experiment 2, 36 naïve listeners compared pairs of widely differing correlations in a 3I-2AFC task with feedback and 2-ERB notches. For 12 listeners who displayed a substantial sensitivity (mean d′ > 1), the four comparisons involving zero gave large d′ values, but comparison of large positive and negative correlations generally gave d′ values of less than one. Only one listener achieved a d′ greater than one for discriminating correlations of 0.8 and −1, indicating that this near-maximal change in interaural correlation is remarkably difficult to detect.