中文摘要：

针对传统压缩感知（CS）方法抗噪性能差的问题，提出了一种新的基于压缩感知的语音增强识别系统。该系统在用正交匹配追踪（OMP）算法重构语音信号时设定相关度闽值和语音恢复阈值，并对迭代算法进行改进，不仅有效恢复了纯净语音信号，实现了语音增强，并且减少了重构的计算量；再将重构恢复的信号通过Gammatone滤波器组提取特征参数GFCC，并在高斯混合模型中匹配。仿真实验表明，将这种方法应用于声纹识别系统，系统的识别率及鲁棒性都有明显提高。

英文摘要：

In view of the poor anti-noise performance of traditional compressive sensing, this paper proposed a novel speech enhancement recognition system based on compressive sensing. The proposed system set similarity threshold and speech signal recovery threshold in orthogonal matching persuit （OMP）algorithm, and improved the iterative algorithm, which not only re- stored the enhanced speech signal, but also reduced the amount of calculation. Then it extracted feature parameters GFCC of the enhanced speech signal by Gammatone filter bank, and matched the best result in Gaussian mixture model. Simulation experi- ments show that this method obviously improves the recognition rate and robustness in speaker recognition system.