Talker and gender effects in induced, simulated, and perceived stress in speech

Abstract

Prior modeling on the speech correlates of psychological stress have typically reported averaged data in which the problem of quantifying degree of stress was not adequately addressed. Harnsberger, Kahan, and Hollien (2006) reported a speech database in which stress was quantified both with (two) physiological measures and (two) self-report scales. In this study, these materials were assessed in two stress perception experiments, involving ratings and classification, and the subsequent analysis was organized by gender and individual talker. Several observations were made. First, both induced and simulated stress samples were perceptibly different from baseline, unstressed samples. Second, up to one third of talkers in a given gender group differed from the overall trend, showing instead a propensity to be rated as less stressed during high degrees of physiological arousal (reflected also in self-report scales); essentially, sounding calmer under stress than the baseline condition. Third, female talkers showed the expected positive correlation between the degree of physiological shift under stress versus the shift in self-reported stress ratings while, in male subjects, higher degrees of physiological arousal were underreported in self-report (a “tough guy” response). Finally, acoustic modeling efforts were conducted by gender and stress response type, for induced, simulated, and perceived stress.