Computer Engineering and Applications ›› 2015, Vol. 51 ›› Issue (15): 216-220.

Previous Articles     Next Articles

Method in noisy signal compressed sensing for speech de-noising

GU Peng1, 2, ZHU Junhua1,2, DING Fei1,2   

  1. 1.College of Telecommunications and Information Engineering, Nanjing University of Posts and Telecommunications, Nanjing 210003, China
    2.Key Lab of Broadband Wireless Communication and Sensor Network Technology, Ministry of Education, Nanjing 210003, China
  • Online:2015-08-01 Published:2015-08-14

一种含噪压缩感知的语音消噪方法

谷  鹏1,2,朱俊华1,2,丁  飞1,2   

  1. 1.南京邮电大学 通信与信息工程学院,南京 210003
    2.宽带无线通信与传感网技术教育部重点实验室,南京 210003

Abstract: To solve the problem of poor reconstruction performance for compressed sensing of speech degraded by white Gaussian noise, a method using wavelet threshold de-noising algorithm based on row echelon matrix measurement sequences is proposed. In addition, when the noise is small, the reconstruction performance decays as a result of the inaccurate estimation of the standard deviations of some frames. To solve this problem, an?improved scheme using one step smoothing for noise estimation is also put forward, which improves the reconstruction results greatly for high SNR. It is demonstrated that by using compressed sensing, this method not only reduces the sampling rate, but also achieves better performance than the traditional speech de-noising methods.

Key words: speech compressed sensing, wavelet threshold de-noising, noise estimation, smoothing, speech enhancement

摘要: 针对含白噪声语音信号采用压缩感知时重构效果差的问题,提出了一种对CS投影矩阵为行阶梯矩阵下的观测序列进行小波阈消噪的方法。针对该方法在噪声较小时,由于对某些帧的观测序列的噪声标准差估计误差过大,从而导致的重构性能衰减严重的问题,提出了一阶平滑的噪声估计的改进方案,改善了高信噪比时的重构效果。仿真显示,该方法通过压缩感知降低采样率的同时,实现了语音增强,并且性能也优于传统语音消噪方法。

关键词: 语音压缩感知, 小波阈值去噪, 噪声估计, 平滑, 语音增强