Human Emotional States Classification Based upon Changes in Speech Production Features in Vowel Regions

Speech signal, that is produced by the human speech production system, carries emotions that the humans can perceive easily. In this paper, we aim to classify the four basic emotions, namely happy, anger, fear and neutral, by analyzing changes in the speech production features in the vowel regions of speech signal. A Telugu emotional speech database is used. Changes in two production features, the instantaneous fundamental frequency (F0) as the source feature, and first three Formants (F1, F2, and F3) as the filter (system) features, are examined in each case. These features are derived from the speech signal using the signal processing methods, i.e., linear prediction (LP) residual and LP spectrum, respectively. The features are examined in the speech segments of five Telugu vowels that have corresponding English vowels, i.e., /a/, /e/, /i/, /o/, and /u/. These vowel regions in the speech signal are detected manually. Further, the classification of emotional states is carried out using a Support Vector Machine (SVM) classifier. The results indicate that in the case of anger emotional state for both male and female speakers, the vowels /a/ and /e/ have higher mean F0 value, as compared to mean F0 for happy, fear and neutral states. Also, the classification accuracy of SVM classifier is observed to be highest for happy emotional state, and lowest for fear emotional state. This insight should be helpful in developing other diverse applications on emotional speech.