基于线性预测的自适应语音压缩感知  

Frame-based Adaptive Compressed Sensing of Speech Signal with LP Analysis

作　　者：赵翠[1] 周遥[1] 毛鑫萍[1] 马永阳[1] 

机构地区：[1]浙江工业大学信息工程学院,浙江杭州310023

出　　处：《杭州电子科技大学学报（自然科学版）》2012年第4期13-16,共4页Journal of Hangzhou Dianzi University：Natural Sciences

基　　金：国家自然科学基金资助项目(6087211);中国博士后基金资助项目(M511901)

摘　　要：该文介绍了压缩感知在语音信号处理中的应用,利用语音信号经过线性预测分析,其残差信号近似认为是稀疏的特性,因此语音信号的稀疏变换基可由线性预测系数构成。同时根据语音帧之间的相关性,该文提出了自适应LP基。实验结果表明该文构造的自适应LP基与非自适应LP基相比,减少了传输量,同时保持重构性能基本一致,且优于DCT基。This paper focuses on the application ear prediction（LP） analysis, speech signals are of compressed sensing（CS） to nature speech signals. In fin- regarded as sparse in the LP basis made by its own set of LP coefficients. Meanwhile, due to the short-time stationary property of speech signals, some successive frames are strong coherent. In this case, the current frame can use the previous＇ LP basis. Therefore adaptive LP ba- sis is proposed. The experimental results show that the reconstruction performance by our proposed adaptive method is similar to that by non-adaptive way, but it is more economic on bandwidth and has better perform- ance than that using DCT basis.

关 键 词：压缩感知 线性预测 稀疏变换基 语音信号 自适应 

分 类 号：TN912[电子电信—通信与信息系统]