The Roles of Pitch Contours in Differentiating Anger and Joy in Speech (original) (raw)

2007, … journal of signal …

The present study is an attempt to use parameters that specify detailed pitch contours to improve the performance of classifying joy and anger emotions in speech. Three parameters, pitch range, speed of pitch change and peak alignment, were taken from accented and unaccented syllables in anger and joy speech samples. We also explore the possibility that alternative strategies of specifying these parameters may be used in the same emotion. The results show that, by using these syllable-based parameters and by allowing for alternative strategies, the performance of differentiating anger and joy was improved by 16%. Moreover, we showed that the new measurements could be used in predicting the pitch contours of accented syllables in anger and joy speech. This suggests the possibility of using these parameters to generate pitch contour related to emotion in speech synthesis. Keywords-emotional speech, multi-strategy classification, pitch contour, predicted line of pitch contour I. INTRODUCTION UMAN speech conveys not only linguistic information but also information about the identity, age, geographical origin, attitude, and emotional state of the speaker. The nonlinguistic information is as important as linguistic information for human communication. In this work we concentrate on the emotional state embedded in human speech signal. Emotional speech recognition and synthesis play important roles in many applications. The direct uses include human machine interaction, entertainment, and business (call center). For indirect uses, the recognition of emotional features in speech can improve the performance of speech recognition systems [1]. Psychologists have been representing emotional state on the activation-evaluation space [2]. These values represent the degree of arousal and pleasantness, respectively. The experimental results from previous works suggest that some acoustic features are associated with general characteristics of Manuscript