Research And Implementation Of Speech Emotion Recognition Techniques In Smart Home

Since the first computer born in this world, people's life has changed greatly. During the past half century, Smart Home has been becoming a reality with the process of information technology and the pursuit of high quality life. Especially, the popularization of Internet, and the rapid development of the consequent artificial intelligence, affective computing and the Internet of Things provide more reference for the implementation of Smart Home. However, people would want emotional communication when they meet the material. So far, emotion recognition technology rarely applies to Smart Home. This thesis analyzes and studies the related problems of necessity and possibility of the implementation of speech emotion recognition technology and the method of implementing Smart Home in detail. The mains contributions are as follows:Firstly, the definition and classification of emotions are introduced, and five main emotions which are recognized by speech emotion recognition in Smart Home are proposed. Then the thesis analyzes the way of obtaining data, discusses the related problems of obtaining speech data, for example the characters of speech and the ways of obtaining data on this premise, proposes three rules about designing speech text and make sure that the way of hybrid recording is a suitable method for Smart Home. It also describes the design and the establishing process of emotion speech database for Smart Home in detail.Secondly, the related problems of feature extraction in speech emotion recognition are studied in this thesis. Starting from the definition of feature extraction, this thesis expounds that feature extraction is necessary for recognition process, and analyzes the characteristics of speech signal. With the specific circumstances of Smart Home, it needs the treated speech signal close to the character of human auditory and including enough information. Mel Frequency Cepstral Coefficient(MFCC) are chose as feature parameter in the thesis and the process of MFCC is introduced in detail.Thirdly, this dissertation studies the problem of feature matching in speech emotion recognition for Smart Home. At first, it expounds the general concept of feature matching. Combined with Smart Home application-specific scenarios, this thesis analyzes and realizes Vector Quantization in detail which is suitable for Smart Home with a small and fixed number of users. Next, feature parameter which has been extracted is applied to Vector Quantization to validate, and the results of recognition are analyzed. Through the analysis, Vector Quantization makes the rate of error recognition expand. To solve this problem, this thesis puts forward Decision based Vector Quantization(DVQ),validates and analyzes these two algorithms. Last, the result is that the recognition rate of Decision based Vector Quantization is better than that of Vector Quantization, and the error recognition rate of five main emotions is declined. This proves the validity of Decision based Vector Quantization.