Toward a social signaling framework : activity and emphasis in speech

Abstract

Language is not the only form of verbal communication. Loudness, pitch, speaking rate, and other non-linguistic speech features are crucial aspects of human spoken interaction. In this thesis, we separate these speech features into two categories -- vocal Activity and vocal Emphasis -- and propose a framework for classifying high-level social behavior according to those metrics. We present experiments showing that non-linguistic speech analysis alone can account for appreciable portions of social phenomena. We report statistically significant results in measuring the persuasiveness of pitches, the effectiveness of customer service representatives, and the severity of depression. Effect sizes of these studies explain up to 60% of the sample variances and yield binary decision accuracies nearing 90%.