行阶梯观测矩阵、对偶仿射尺度内点重构算法下的语音压缩感知

基于语音信号在离散余弦域上的近似稀疏性,针对采用随机高斯观测矩阵及线性规划方法进行语音压缩感知与重构时,重构零(近似零)系数定位能力差而导致重构效果不好的缺点,本文提出一种新的行阶梯矩阵做观测矩阵,用对偶仿射尺度内点重构算法对语音进行压缩感知与重构,并对该算法下的重构性能进行理论分析.语音压缩感知仿真结果表明,在离散余弦基下,压缩比(观测序列与原始序列样值数之比)为1∶4时,行阶梯观测矩阵下的平均重构信噪比比随机高斯观测矩阵下提高9.73dB,平均MOS分比随机高斯观测矩阵下提高1.22分.

Compressed Sensing of Speech Signal Based on Row Echelon Measurement Matrix and Dual Affine Scaling Interior Point Reconstruction Method

Based on the approximate sparsity of speech signal in discrete cosine basis,this paper proposes a new algorithm of compressed sensing of speech signal based on special row echelon measurement matrix and dual affine scaling interior point reconstruction method.This algorithm can resolve the problem of inaccuracy of location of reconstruction coefficient which is zero or nearly zero of compressed sensing based on Gaussian measurement matrix and linear programming to some extent.The reconstruction performance of this algorithm is analyzed theoretically.The simulation results of compressed sensing of speech signal show when the reduction ratio(the ratio of numbers of measurements and original samples) is 1∶4 based on the discrete cosine basis,the average SNR of reconstruction signal based on the special row echelon measurement matrix is 9.73 dB higher than the Gaussian measurement matrix,and the average MOS score of reconstruction signal based on the special row echelon measurement matrix is 1.22 higher than the Gaussian measurement matrix.