Abstract: Based on the compressed sensing theory, it proposes a pre-treatment for the sparsity of transform-domain by zeroing the value below the threshold in the DCT domain. It builds the circulant measurement matrix in three ways instead of Gaussian random matrix, and proves the non-coherence between measurement matrix and DCT base. It uses the OMP method to recover the signal. Simulation experimental result demonstrates that after pre-treatment using circulant measurement matrix has lower restruction error in different compression rate. While analyzing the SNR of each frame, it guarantees a low compress rate and has a good score in PESQ.

Key words: speech compressed sensing, Discrete Cosine Transform（DCT） sparse pre-treatment, circulant measurement, Orthogonal Matching Pursuit（OMP）

摘要： 基于压缩感知原理提出将语音信号DCT域上的小系数在一定阈值下置零预处理来改善变换域稀疏性；用三种方法构造循环观测矩阵作为观测矩阵来代替高斯随机矩阵，并证明了构造的观测矩阵与DCT基之间的非相关性；利用OMP正交匹配追踪方法对观测信号进行恢复。仿真实验结果表明，预处理后使用循环观测在不同压缩率下有更低的重构误差，同时分析各帧信噪比情况保证在比较低的压缩率下仍能得到良好的主观评估。

关键词: 语音压缩感知, 离散余弦变换（DCT）域稀疏预处理, 循环观测, 正交匹配追踪（OMP）