基于Sugeno测度的动态不可加双矢量量化说话人识别研究  

Speaker recognition research of dynamic nonadditive double vector quantization based on sugeno measure

作　　者：田景峰[1] 赵振东[2] 胡喜梅 

机构地区：[1]华北电力大学科技学院,河北保定071051 [2]华北电力大学电气与电子工程学院,河北保定071003 [3]中国移动通信集团河北有限公司保定分公司,河北保定071000

出　　处：《华北电力大学学报（自然科学版）》2010年第6期106-109,共4页Journal of North China Electric Power University：Natural Science Edition

基　　金：华北电力大学校内科研基金资助项目(200911033)

摘　　要：针对双矢量量化方法中语音的静态特征和动态特征的权重不满足可加性的情况,提出了一种新的说话人识别方法——基于Sugeno测度的动态不可加双矢量量化说话人识别方法。该方法在Sugeno测度空间上将说话人语音的静态特征和动态特征用Sugeno测度进行动态融合。然后,在噪声环境下研究了该方法的识别效果,找到了噪声环境下语音的静态特征和动态特征参数的较优的权重组合。实验结果表明,与双矢量量化识别方法相比,该方法可以使识别率明显提高。该方法为研究各类语音特征参数之间的关系、探寻最优的特征匹配方案提供了一种新的途径。In view of the weight of speech static characteristics and dynamic characteristics does not necessarily satisfy the additive situation in double vector quantization（D-VQ）,a new double vector quantization speaker recognition method—dynamic nonadditive D-VQ speaker recognition model based on Sugeno measure was proposed.This method integrated the speech static characteristics and dynamic characteristics of the speaker with the Sugeno measure on Sugeno measure space.Then,the recognition effect of this method was studied in noisy environment,so that better weight combination of static characteristics and dynamic characteristics was founded in noisy environment.The results of the experiment show that compared with D-VQ method the proposed method can highly improve the recognition rate.The method provides a new way to study the relationship of various speech features and explore the optimal feature matching scheme.

关 键 词：双矢量量化 说话人识别 Sugeno测度 动态不可加双矢量量化 权重组合 

分 类 号：TN912.3[电子电信—通信与信息系统]