Abstract:
A listener's ability to integrate information can be evaluated by presenting multiple samples in the distribution discrimination procedure. To examine frequency discrimination with a 2IFC form of this procedure, each interval contains samples drawn from one of two distributions of signal frequency. The listener indicates which interval contained the sample(s) drawn from the distribution with the higher mean. This experiment evaluated the integration of frequency information in terms of improvement in frequency discrimination for an increasing number of samples (n=1,2,3,4,5,6,8,12,16). Seven listeners discriminated a ``standard'' distribution (means of 400, 565, and 1000 Hz) from each of four ``comparison'' distributions (means of 401, 403, 406, and 414 Hz for the 400-Hz standard; 566.5, 569.5, 572, and 584 Hz for the 565-Hz standard; and 1002, 1005, 1010, and 1020 Hz for the 1000-Hz standard). All samples were 100-ms sinusoids. All conditions were alike to an ideal observer in that the distributions were normal with a standard deviation equal to the difference between the means. The results indicate that integration of frequency information appears constant across different frequencies when initial performance is equated. [Work supported by AFOSR through WPAFB AL/CFBA.]