Speech Emotion Recogonition Research

Emotion recognition from speech has key role in improving human-computer interaction. It will make the computer services more kindly. Based on matlab7.0, this dissertation researches the technologies of speech emotion recognition. Finally a completed recognition system is constructed.In this paper, eight feature parameters which are closely related to the emotion were extracted. Those parameters were unified quantized by Vector Quantization. A discrete HMM recognition model was selected to recognize the five kind of emotions. The recognition rate is been counted. Finally, the paper was been summarized and the further study was prospected. The main work is as follows:First, the human-independence and text-independence speech emotion database was obtained in this paper. There are five kinds of emotions in the corpus data:happy, surprise, anger, grief and neutral.Second, the feature parameters were extracted after preprocessing of the speech signals. There are short-time average energy, short-time average amplitude, short-time zero crossing rate, short-time autocorrelation function, fundamental frequency, LPCC, MFCC and the first formant peak.Finally, the discrete hidden Markov recognition model was used to recognize the five emotions after the extraction of characteristic parameters. Vector quantization method was used to get the discrete parameter.32 VQ code were selected to recognize the speech emotion. Baum-Welch algorithm is used to train the parameters; forward-backward is used to test the recognition result. The average recognition rate of men is 65.46%, The average recognition rate of women is 64.04%.