A measure of internal noise based on sample discrimination of level differences

Abstract

In a sample discrimination task, subjects are presented with stimuli drawn from each of two overlapping distributions and asked to identify the distribution from which the stimuli were drawn. Lutfi [J. Acoust. Soc. Am. 86, 934–944 (1989)] has shown that performance in this task is nearly ideal when the stimuli are single sinusoids differing in level, for distributions with a mean difference of 5 dB and standard deviations of 5 dB. Theoretically, as the mean difference and standard deviation of the distributions are decreased, however, performance should increasingly deviate from ideal, yielding an estimate of internal noise. To test this hypothesis, intensity discrimination was measured for seven subjects’ sinusoids drawn from distributions that had mean differences ranging from 0.1–2.2 dB, in steps of 0.3 dB. The standard deviation of each distribution was always half the mean difference, resulting in uniform ideal performance. Estimates of internal noise obtained by fitting a function to the observed d-prime values agreed with estimates obtained from psychometric functions measured using the same differences in level. Analyses of decisions on individuals trials indicated that the two observation intervals were weighted equally. [Work supported by NIDCD.]