Research And Implementation Of Voiceprint Recognition Techniques In Smart Home

With the progress of science and technology and the popularity of the advanced idea of life, the smart home is gradually becoming reality. Especially in recent years, the astounding advance of Internet, Artificial Intelligence and Biometric Identification Technology, has brought various kinds of technical reference for the development of SH. Through it, we can make the advance of technology contribute to the feeling of mankind. However, not even at home but also abroad, the Biometric Identification has few implementations besides the vital security department. As to the technology of voiceprint recognition, it is really rare to be seen in the field of SH.In this thesis, it is studied deeply that problems of implementation for voiceprint recognition in the field of smart home. The contribution point of this paper is summed up as follows:1. The problem of feature extraction on voiceprint recognition is discussed firstly. Two algorithms of feature extraction have been implemented after detailed analysis of voice print, combining to the specific scene in smart home. The mentioned algorithms are linear prediction coefficient (LPC) and Mel-frequency cepstrum coefficient (MFCC). Through experiment data found LPC algorithm has better time performance than MFCC, but the result file produced by MFCC algorithm has richer feature information than LPC.2. Secondly, the issue of feature matching is studied. According to the different characteristics of the emphasis of the voiceprint, vector quantity (VQ) algorithm is chosen and implemented. At the same time, the thesis presents three kinds of distance algorithms and two ways of clustering algorithms. After integration of the Euclidean distance, City block distance and Chebyshev distance with LBG and K-means algorithms, six kind of combined models are listed. According the comprehensive experiments, optimal combination of feature matching is obtained for voice recognition with small-scale of users.3. Finally, the thesis presents the survey of speech database and speech capture in voice recognition. The speech database is divided into two subsets for training and test. A variety of speech datasets are designed in consideration of the influence of speaker’s gender and speech length, etc.