Relative amplitude based features for emotion detection from speech

Emotion detection from speech has been realized to provide benefits for more natural human-machine interaction. To detect the emotion from speech signal, an abundantly long continuous speech segment is needed. This paper proposed a navel approach for emotion detection based on relative amplitude of speech signal. Relative amplitude reduces bias of glottis mutation of speech wave amplitude and obtains a normalized measure without concern of information from being distinct in feature. RBFC approach is used for segmentation of speech signal and the results are compared with other voiced segmentation approaches. Berlin emotional speech database is used for experimental purpose. The results show that accurate emotion recognition is obtained with optimum length of emotional speech. The RBFC features generate more accurate results than the other methods.