基于改进循环观测的线性预测语音压缩感知  

Linear Predictive Speech Compressed Sensing Based on Improved Circulant Observation

作　　者：徐皓波 于凤芹[1] 

机构地区：[1]江南大学物联网工程学院,江苏无锡214122

出　　处：《计算机工程》2014年第11期278-281,共4页Computer Engineering

基　　金：国家自然科学基金资助项目(61075008)

摘　　要：利用语音信号线性预测残差的稀疏性特点可对语音信号进行压缩感知,但需要信号的线性预测系数来构造稀疏变换矩阵,从而增加预测系数传输的数据量。为此,提出将线性预测系数存入对角阵向量中构造循环矩阵,由此得到循环观测矩阵,再对语音信号进行观测。提取该循环矩阵中的线性预测系数构造残差域稀疏变换矩阵,利用正交匹配追踪算法对观测信号进行重构。仿真实验结果表明,与传统线性预测方法相比,该方法减少了3.9%以上的数据量,且比将高斯随机矩阵作为观测矩阵的方法具有更高的帧平均重构信噪比。The sparsity of the linear predictive residual of speech signal can be used in the speech compressed sensing, but needs the predictive coefficients of the signal to build sparse transformational matrix increasing the data. This paper proposes to save the linear predictive coefficients into diagonal matrix in order to build circulant matrix,and measures the speech signal in circulant way, extracts the linear predictive coefficients from circulant matrix to build sparse transformational matrix in residual domain, and reconstructs the speech with Orthogonal Matching Pursuit （ OMP ） algorithm. Simulation experimental result shows that, using circulant measure built by predictive coefficients decreases 3. 9% data more than the original linear predictive method,and has higher reconstruction signal to noise ratio per-frame than the Gaussian random matrix as measure matrix.

关 键 词：线性预测 压缩感知 循环观测 残差域稀疏变换 正交匹配追踪 重构信噪比 

分 类 号：TN911.7[电子电信—通信与信息系统]