一种基于压缩感知的说话人识别参数分析  

Parameter of Speaker Recognition Based on Compressed Sensing

作　　者：潘海琦 杨震[2] 徐珑婷 朱俊华[1] 

机构地区：[1]南京邮电大学通信与信息工程学院,南京210003 [2]南京邮电大学"宽带无线通信与传感网技术"教育部重点实验室,南京210003

出　　处：《数据采集与处理》2015年第2期399-407,共9页Journal of Data Acquisition and Processing

基　　金：国家自然科学基金(61271335)资助项目;江苏省普通高校研究生科研创新计划(CXZZ13-0488)资助项目

摘　　要：本文为在传统的说话人识别理论研究中"较少的特征参数量不能与较高的识别率共存"的难题找到了一种解决方案。本文基于压缩感知的理论,利用行阶梯观测矩阵进行信号的投影,改变了传统的梅尔频率倒谱系数(Mel-frequency cepstral coefficient,MFCC)参数,从而提出了一种新的识别参数CSMFCC(Compressed sensing-MFCC)。该参数不仅使得参数存储量降低到少于原存储量的1/n(n为行阶梯观测矩阵的压缩比),而且明显提高了系统的鲁棒性。通过仿真实验证明了当压缩比n为4时,平均识别率能够提高到96%以上。A solution is proposed to deal with the problem that″less number of features cannot coexist with higher recognition rate″in the traditional theory of speaker recognition.Ladder observation matrix projection is used to change the traditional Mel-frequency cepstral coefficient（MFCC）parameters based on compressed sensing theory,presenting a new recognition parameters named compressed sensing MFCC（CS-MFCC）parameters.These parameters make storage capacity decrease to less than 1/nof the original,here nis the compression ratio of the line ladder matrix,and also greatly increase the robustness of the system.Furthermore simulation results prove that when nis 4,the recognition rate increases to96% above.

关 键 词：说话人识别 压缩感知 识别率 CS-MFCC 鲁棒性 

分 类 号：TN912.3[电子电信—通信与信息系统]