基于DSP开发板的语音识别系统的研究  

Study of Speech Recognition System based on DSP

作　　者：杨海燕[1] 景新幸[1] 曾招华[1] 

机构地区：[1]桂林电子科技大学信息与通信学院,广西桂林541004

出　　处：《计算机测量与控制》2013年第1期210-212,220,共4页Computer Measurement &Control

基　　金：国家自然科学基金资助项目(609661002)

摘　　要：文章以SVM作为语音识别核心算法,以TMS320C6713DSK作为硬件平台,选择相应的语音识别算法,利用SVM和DSP各自的优势,设计了一款非特定人、小词汇量的连续实时语音识别系统;并在Matlab平台上进行了实验仿真,然后在TMS320C6713DSK开发板上进行论证;实际系统前端采用MFCC特征参数、并用遗传算法与矢量量化混合算法对其进行聚类得到优化码本,所采用种群数为80,码本长度为16,然后将所得码本作为SVM模式训练和识别算法的输入;结果表明,该系统在满足系统实时性的同时也提高了其正识率,具有一定的实用性。In this paper, SVM is used as the core algorithm for speech recognition system, choosing TMS320C6713 DSK as the hardware platform, combining the advantages of SVM and DSP, It is designed for a small--vocabulary, speaker--Independent continuous real -time speech recognition system. Then the prepared algorithms are simulated in Matlab. Finally, the speech recognition system is simulated and demonstrated on DSK platform. The front of the actual system uses MFCC feature parameters, and the hybrid algorithm of genetic algorithm and vector quantizationclustering, the code book has been optimized. Among the population be 80, yards of this length be 16, and then the codebook is used as the input of SVM model. The results show that the system not only meet the demand of a real-- time system, but also in creased its positive recognition rate. The system developed in this paper are practical and worthy of using abroad.

关 键 词：语音识别 MFCC DSP SVM 

分 类 号：TN912.3[电子电信—通信与信息系统]