SVM And HMM Combination Of Design And Implementation Of Chinese Speech Syllable Recognition Algorithm

Speech recognition is the beginning that people move towards the comprehensive intelligence, and people hope to apply voice instead of manual operation and use the uniqueness of the voice to do authentication. The machines can recognize the voice by the forms of order and code. In order to achieve this objective, scientists have done a lot of research by means of pattern recognition, probability and statistics, distinguishing the authenticity; at same time they solved many problems, having achieved a voice-activated phone dialing, typing dictation technology and voice-activated access control, speech-driven and a series of technologies.In this thesis, we mainly discuss that the Hidden Markov Model apply on the Chinese speech recognition, then it introduces the Chinese speech recognition system and 3D facial lip animated by Speech-driven, the main work is as follows:1. First, a common statistical model, Hidden Markov Model, is researched and analyzed in the thesis. Based on this model, the thesis presents a new Chinese-language speech recognition algorithm. In the circumstance of the laboratory, we use MATLAB to prove the feasibility and effectiveness of this algorithm well. The recognition rate can reach 81.6% based on the algorithm.2. For raised the recognition rate, I researched to improve the algorithm. After reading and researching the excellent papers and materials at home and abroad, a support vector machine is used to do classification well, while the Hidden Markov Model is very effective to do word recognition. Hence the thesis will combine the advantages of both to improve algorithm. The recognition rate can reach 93.5% by simulation of MATLAB. The experimental results show that the recognition rate of the improved algorithm has been greatly improved.3. Following the preceding Chinese-language speech recognition algorithm based on SVM and HMM, the thesis designs and implements a Chinese Speech syllable recognition system based on SVM and HMM. Based on the system, first, the system has done a more comprehensive need analysis, have given an overall design of the system and system module into the division. The thesis introduces the system development process from the system needs analysis, overall design, features and system module in details. The system adopts C / S mode to develop itself by User's putting forward request, the server answering the request and its returning to recognition results.4. Speech-driven 3D facial lip is described in the last part of the thesis. Based on the previous algorithm, we propose Chinese speech syllable to drive the lip animation. Since the algorithm is more applicable to word recognition. This thesis gives a brief introduction of the speech segmentation algorithm. The recognition rates of the speech-driven 3D facial lip reached 76.7%. We believe this recognition rate is very satisfied. Finally, we propose the combination of expressions and lip-animation to make 3D face livelier and more realistic.