关于语音个人身份优化识别建模仿真研究

About the Speech Personal Identity Optimization Modeling and Simulation Research

声纹可以代表语音信号中个人的信息特征,是语音个人身份优化识别的关键。进行声纹识别时,必须要从语音信号中提取相关的声纹信息,对声纹信息进行全局分布特性的最优聚类,而传统方法在选取聚类中心点时采用随机选取,不能形成全局分布特性的最优聚类,导致迭代过程陷入局部最优解的缺点,降低了声纹识别的精度。为提高识别精度,对语音个人身份优化识别进行建模仿真,通过小波变换阈值法对语音信号进行去噪处理。利用小波变换方法和MFCC特征参数的提取原理相融合用来提取语音特征信号。通过高斯混合模型算法将提取出的新特征参数DWTC建模,在模式匹配的训练阶段利用EM算法求取参数集,并加入LBG算法求取起始参数值并通过MAP准则实现模式识别。仿真结果表明,改进小波声纹识别中的应用仿真平台有效的设计出能够对说话对象的声纹信息的有效提取,可获取说话对象的声纹信息进行有效辨识。

In order to improve the identification accuracy,the modeling and simulation of optimum personal identification of voice were carried out. The voice signals were denoised with wavelet transform threshold method,and the extraction principle integrated wavelet transform method and the MFCC feature parameter were used to extract the voice characteristic signal. By the Gauss mixture model algorithm,the new extracted feature parameter DWTC was modeled. In the training phase of pattern matching,the parameter set was obtained by EM algorithm and the LBG algorithm was added to obtain the initial parameter value,and the pattern recognition was realized by MAP criterion.The simulation results show that the simulation platform based on improved wavelet voiceprint identification can effectively design the method which can effectively extract the voiceprint information of the speaking object,and obtain the voiceprint information of speaking object to make identification.