ASA 125th Meeting Ottawa 1993 May

4pSP23. Listener performance under variations in signal content.

Martin Schultz

Commun. Disord. and Sci., Southern Illinois Univ., Carbondale, IL 62901

Tan Kah

Southern Illinois Univ., Carbondale, IL 62901

Signal detection theory holds that performance can be modeled under conditions of physical specification of signals and systems, with no attention given to signal informational content. By contrast, Shannon [Bell Syst. Tech. J. 30, 50--64 (1951)] proposed that performance is open to coding influences and, therefore, may vary with signal content. This study examined performance of subjects for three-element sequences wherein the content varied systematically from words, to nonword, vocal system outputs, to speech-patterned noises. Signals formed a twenty sequences closed set, randomly selected and delivered with equal probabilities. Subjects were trained prior to experimental runs to a 90% criterion but probably not to asymptotic performance, using trial-by-trial feedback. All signal elements were represented by multiple tokens and all signals were of equal power and duration. Training was accomplished in the ``clear,'' signal/noise ratio was an experimental variable and both training and experimental conditions were self-paced. Subjects were paid for performance throughout. Results are discussed from several theoretical perspectives.