Recognition of Practical Emotion from Elicited Speech

Abstract

To meet the real world challenges for speech emotion recognition, four emotions for practical use were studied for evaluation of work ability. Elicited speech corpus was collected in a psychology experiment to provide trustable emotion data, acoustic features related to arousal and valence dimensions were selected specially for the practical emotions and a re-compositive GMM method was used for the classification. Twenty best acoustic features were achieved and a satisfactory recognition rate was observed. The results suggest that the classification of four practical emotions is successful and the elicited speech corpus is suitable for building a practical emotion recognition system for the real world challenges.