Research And Implementation Of Speaker Recognition System Based On Haar Wavelet Packet Analysis

With the development of information processing and computertechnology,Today the development of speaking and speaker recognition technologyhas been in various fields are development and application of rapid, and voiceprintrecognition as an important branch of speech recognition technology, it is animportant research direction in speech recognition technology. Voiceprint recognitiontechnology has very broadly application prospects in many fields, judicialreconnaissance, electronic commerce, military communication and so on. involveslinguistics, speech signal processing, pattern recognition and machine learning fieldsof knowledge, and this paper mainly discusses the realization of voiceprintrecognition system architecture.In this paper, analysis and summary of existing speech recognition systems andthe existing main problems, the design and implementation of a new voiceprintrecognition system.Deeply research the key technology of voice feature extraction,present the improvement of Mel-scale Frequency Cepstral Coefficients algorithmusing haar wavelet packet analysis. this algorithm could reduce amount of calculationof voice feature extraction and improve voiceprint recognition system accuracy.Build the mixed Gauss model voiceprint feature (GMM) library of voiceprintrecognition using K cluster analysis method. Relate a solution to solve that, blindclassification is not accurate and lead to mixed Gauss model is not accurate. Adoptdouble threshold endpoint detection algorithm of speech signal processing, analysisand gives the double threshold selection scheme of each threshold endpoint detectionalgorithm for the reference. The MATLAB simulation experiment show that thesystem has the characteristics of fast feature extraction speed and better recognitionaccuracy, reach the expected goal.