Research And Implementation Of Gaussian Mixture Model-based Speech Emotion Recognition

As the use of computer has been becoming more and more important, we hope to enhance the humanity and the intelligence of computer and realize more harmonious interaction between human and machine.As the most natural mode of human communication, speech contains lots of emotional information of the speaker, so how to recognize speakers' emotional state has been paid more attention by researchers. In order to accomplish that, computers should finish affective computing. Recent studies on speech emotion recognition used different modals, emotion-speech databases and speech emotion features, so the results are different. Especially researches in Chinese speech emotion recognition need more development.First, a Chinese emotional speech database has been constructed for experiments. Sentences, which don't contain any emotion themselves, were read with 7 emotions. We selected the sentences that emotions were well expressed for our analysis and experiments. Then through observing and analyzing, we selected and defined the features (pitch, resonance, energy, etc.) which are the most important in distinguishing emotions.Based on the selected features and analysis of algorithms used in speech emotion recognition, the paper selected GMM as the recognition algorithm to recognize speech emotion. We studied the training and recognition algorithms of GMM, built GMMs for 7 emotions. By having analyzed the emotion recognition experiments, the pitch related features are useful to distinguish sad emotion state. After considering the resonance, we achieved the improved recognition rates for the 7 emotions. Experiment results also show that speed and average energy are discriminant for the 7 emotions.