Informational masking in profile analysis: comparing ideal and human observers

Predictions from an ideal observer model are compared with human thresholds for two profile analysis tasks. Past work has shown that ideal observer models reasonably account for human thresholds when the profile components are fixed in frequency and amplitude. Randomly varying the frequencies of the tones making up the profile leads to higher thresholds. Owing in part to large interobserver variation, the ideal observer model is not successful in accounting for the pattern of psychophysical thresholds associated with increases in frequency uncertainty. The ideal observer also fails to account for the results of a recent profile analysis experiment in which amplitude randomization was studied [Lentz JJ, Richards VM: J. Acoust. Soc. Am. 102: 535-541, 1998]. Overall, the ideal observer predicts smaller effects of uncertainty on thresholds than are observed in psychophysical experiments.