A correlational study of human emotions and voice properties: exploring positive and negative affects using time series and frequency domain measurements

Prior research, mostly with clinical patients, has shown that there are cues in the\ud human voice that are correlated with various emotional states. The purpose of this\ud study is to explore more broadly correlations between eight positive and negative\ud emotion states, and eight speech measures. The extension and application of these\ud analyses may provide yet an additional objective, discriminating measure of human\ud emotion. The hypotheses were that at least several of the moods would be predictable\ud from the voice measures, and that previous correlations between voice and mood\ud would be replicated (that negative affects would be characterized by the voice being\ud slow, lower frequency, and less loud). Significant correlations were found between\ud most of the emotions and the voice properties measured. The regression analysis of\ud emotions (criteria) produced significant regression coefficients with voice\ud measurements (predictors); the resulting fingerprint of voice parameters can be\ud employed to identify as well as discriminate between emotions. This study revealed\ud that the use of prosody (time domain) parameters, and the telephone as a vehicle,\ud could be an important part of emotion-detection from voice. Several limitations and\ud future directions to this study are discussed