TEO能量与Mel倒谱混合参数应用于说话人识别  

TEO Energy and Mel Cepstrum Mixed Parameters Used in Speaker Recognition

作　　者：杨瑞田 周萍[1] 杨青[1] 

机构地区：[1]桂林电子科技大学电子工程与自动化学院,广西桂林541004

出　　处：《计算机仿真》2017年第8期215-219,264,共6页Computer Simulation

基　　金：广西自然科学基金(2012GXNSFAA053221;2014GXNSFA A118353);广西研究生教育创新计划资助项目(YCSZ2015152);国家自然科学基金(61363005;61462017)

摘　　要：特征提取是说话人识别中非常重要的一个环节,特征提取的结果直接影响系统的识别结果。提出一种将TEO与MFCC及其衍生参数结合的方法,将本文提取的特征参数与传统的MFCC,WMFCC与△MFCC通过GMM-UBM与SVM模型得出结果并比较。并在不同环境下的进行实验,对算法进行了仿真实现。实验结果表明,在相同噪声背景不同信噪比时与相同信噪比不同的噪声背景这两种情况,提出的方法均得到了较好的结果,在检测纯语音数据时,对融合算法进行仿真实现,识别率也得到了提高。The extraction of characteristic is a very important part in speaker recognition. The resuk of the extrac- tion directly affects the recognition result of the system. In this paper, a method is proposed which combines TEO with MFCC and its derivative parameters, and the feature parameters extracted from the method were compared the results with the traditional MFCC, WMFCC and A MFCC though the GMM - SVM and SVM models. And the algorithm was simulated and implemented through the experiments in different environments. In the two cases of same noise back- ground with different SNR and same SNR with different noise background, the experimental results show that the pro- posed method has acquired better results. In the detection of pure speech data, the mix algorithm was simulated, and the recognition rate was improved.

关 键 词：说话人识别 梅尔频率倒谱系数 高斯混合模型-通用背景模型 联合因子分析 

分 类 号：TN912.34[电子电信—通信与信息系统]