Quantifying the Impact of Speech Intelligibility on Task Performance via Physiological Signals

Abstract

Background noise can potentially cause dissatisfaction, stress, and reduced performance in the workplace. One type of background noise comes from background discussions. Many studies in the literature exist on the effects of speech intelligibility, a common measure of speech comprehension. Presently, no unitary framework exists for describing the relationship between stress and performance; furthermore, the exact constitution of stress response in terms of physiological and subjectively measured characteristics is not yet determined.

This thesis investigates the impact of various speech intelligibility conditions on cognitive task performance in a simulated office environment using electrodermal activity (EDA) and heart rate (HR) physiological signals. It was hypothesized that increase in distraction causes increase in stress, which causes decrease in performance, and thus distraction and performance are inversely related. A user study was conducted in a simulated office setting where participants (N = 24, N = 29) performed different cognitive tasks listening to audio of variable speech intelligibility conditions. Data collected from participants included EDA and HR signals and self-reported subjective ratings during the experiment, and task performance scores. Correlations were computed to help assess conformance of the data to the hypothesized relationships. Physiological signals and self-reported subjective ratings were hypothesized as indicators of stress; correlations were computed to help determine how stress could be measured between them. This thesis also aims to determine which intelligibility conditions have a stronger effect on stress and performance, and how certain findings are affected by the type of cognitive task and by experimental design variations, namely, (1) Within-Subjects design and (2) Between-Groups (randomized) design.

The following were found from the analysis results. Distraction was not directly related to stress in either experimental design. Stress was inversely related to performance in both designs. Distraction was inversely related to performance in the second design. Physiological signals and self-reported stress ratings were a more accurate stress indicator in the first and second design, respectively. The high intelligibility condition overall affected performance strongest in the second ii design but not stress in either design. Overall the Between-Groups (randomized) design yielded results more in line with our research hypotheses than the Within-Subjects design.