Statistical feature based child emotion analysis

Abstract

To characterize any speech signal, features plays an important role as a parameter to best describe a particular speaker from his/her voice. In emotional speech recognition system prosodic and spectral features provide a significant detecting parameter in differentiating various classes of emotions as these features closely resembles human vocal tract system. The statistical properties of these features vary with different emotional speech utterances and due to change in ascent, speaking style and language of the speaker. In this paper, an attempt is made taking into account these facts to categorize various classes of emotional speech. The result is promising in detecting four classes of emotions as angry, fear, sad and surprise of children database generated by us.