Affective State and Voice: Reproducibility and Sensitivity of Speech Parameters

Abstract

SummaryIn 1986, we started an extensive investigation into the nonverbal information transmitted in human speech. As part of this investigation, we had designed a study of healthy subjects -stratified according to sex, age and education - in order to gain valuable insights into the “typical” nonverbal activities of a speaker. The specific design of this study with 3 different types of texts and 2 repeated measurements at an interval of 14 days was used to test the stability of speech parameters over time as well as their sensitivity to form and content of text. In a first approach, we directed our efforts towards a set of major scalar quantities which earlier investigators had described as playing a key role in modelling affect in terms of directly measurable speech parameters.Based on a sample of 97 persons and with the help of a fully computerized digital approach to speech processing, we found that the four major speech parameters “mean vocal pitch”, “time duration of pauses”, “time duration of utterances”, and “energy” are sufficiently reproducible under constant experimental conditions, whereas some significant differences showed up with regard to the sensitivity of these parameters to form and content of the spoken text. No substantial impact originated from the external factors sex, age and education, except for the tonal shift of one octave between males and females.On the other hand, the overwhelming variety of individual voice patterns obviously excludes any simple model (which applies in general) of the nonverbal component of human speech. However, our analyses offered some important clues as to how appropriate models may be derived by means of adaptive procedures.