The ideal observer: The forgotten second half of signal detection theory

Abstract

In his seminal text “Signal Detection Theory and Psychophysics” (1966), coauthored with John Swets, Dave Green identifies two parts to the theory. The first entails the psychophysical methodology and procedures required to separate sensory and decision processes in detection, the second compares human performance to that of a theoretical ideal observer that optimizes decisions based on the statistical properties of signals and the demands of the detection task. The first part has received far wider application in psychophysics, but there are demonstrable advantages to applying both. In this talk I will give examples from work in our lab where the analysis of ideal observers framed in the methodology has been used to (1) give mathematical meaning to the constructs of stimulus uncertainty and similarity in informational masking; (2) determine sensory constraints on the use of invariant acoustic cues in sound source identification; and (3) isolate a fixed effect of listeners in multi-talker speech segregation across experiments involving vastly different stimuli and psychophysical tasks. [Work supported by NIDCD grant R01 DC001262.]