基于语音信号的情绪识别研究机械电子工程专业论文.docxVIP

哈尔 哈尔滨工业大学工学硕士学位论文 - - PAGE VI - Abstract With the advancement of human-machine interaction, lack of emotional has become a greater barrier, so “affective computing” science appeared. Speech emotion recognition is a key branch of affective computing. Speech emotion recognition has been used in all fields of interaction. This paper studied the emotions in Chinese speech mainly. Gaussian Mixture Model and Expectation Maximum algorithm were applied to intelligent car controlled by the mood of voice. Intelligent level was enhanced in the process of human-machine interaction. Firstly, human beings’ pronunciation mechanism was analyzed and several characteristics of pronunciation were summarized. Voice database was set up by means of inducing record combined with the principle of recording scripts selection and this research topic. Characteristics parameters were extracted according to the relationship between pronunciation and emotion. Center-clipping was applied in pitch detection by autocorrelation function method to accelerate calculation velocity. Secondly, emotion classification was carried out with Gaussian Mixture Model, which was progressed, and logarithm likelihood function was maximized by Expectation Maximum algorithm, resulting to enlarge the range of speech emotion. MATLAB program was written to complete the weights learning and emotion recognition, thus feelings classification in speech signal was realized. At last, the route of the car, whose control panel was developed by Sunplus company and driven by DC driver, was controlled by mood of voice. Experiment results show that, emotion classification was reasonable and effective by means of GMM combining with EM algorithm. It was proposed to improve the safety in the human- computer interaction due to emotion control. Emotion extent of Chinese was explored. Keywords: affective computing; speech emotion recognition; speech characteristic parameters extraction; Gaussian Mixture Model; Expectation Maximum algor