一种快速自适应噪声谱估计方法

Fast adaptive estimation of noise spectrum

摘要: 提出了一种快速自适应的噪声谱估计方法。该方法在Doblinger和Cohen的噪声谱估计方法基础上,提出了改进算法,修正了带噪语音子带中语音存在的判断门限和噪声谱更新公式,估计的噪声谱既能快速适应背景噪声变化,又能保证准确性。实验结果表明,结合文中噪声谱估计方法构成的语音增强系统能有效抑制带噪语音中平稳和非平稳噪声,对于噪声突变的情况也适用。

Abstract: A fast adaptive method for noise spectrum estimation is proposed in this paper.We improve the algorithm of noise spectrum estimation proposed by Doblinger and Cohen.This is achieved by modifying the threshold for judging presence of speech in sub-bands and the formula of noise spectrum update.The estimated noise spectrum can quickly adapt with background noise and keep accuracy at the same time.Results show that,by using this method,a speech enhancement system has good performance both in stationary and non-stationary noise suppression,and also under sudden change of noise.