摘要
针对时域中含噪语音信号较长而无法直接进行奇异值分解以及在分解中无法自适应确定分离阶数的问题,本文提出了将时域语音信号分段后再进行奇异值分解的方法,该方法在分段分解后通过拼接可还原出原信号。为了解决分离阶数选取困难的问题,在分解过程中本文采用均值法及方差法自适应确定分离阶数,仿真实验结果验证了该方法对语音信号去噪的有效性。
Considering that speech signal with noise is long. It can' t use singular value decomposition directly(SVD) and determine the separate order number. This article proposes a method that segments the speech signal firstly and then uses SVD. It can splice the segments to reduction signal using this method. To solve the problem that separate order number is difficult to select.In the process of SVD, separate order number is selected adaptively by averaging and variance method in this article. Simulation results verify the effectiveness of the method on denoising.
关键词
奇异值分解
语音信号
去噪
分离阶数
Singular value decomposition Speech signal Denoising Separate order