Detection of emotional expressions in speech

This paper provides a survey literature survey on the emotion recognition in spoken dialogs and proposes an implementation of such a system using acoustic features. The data corpus contains 322 utterances expressing four emotions such as happy, angry, sad, and fear. 50% of the total data is used for training while the other 50% is used for testing. We use 21 features extracted from our features set in our experiment. The feature vectors are normalized by using Z-score normalization. The mufti-class Support Vector machine (SVM) classifier is used for classification. The result shows that sad is classified with the highest accuracy whereas happy is classified with the least accuracy.