基于最小均方误差估计的改进语音增强算法

在常规谱减法的理论基础上,针对其去噪中存在的问题,提出一种改进算法,对清音和浊音分别采用高频提升和频带加宽的处理方法,使在消除信号噪声的同时尽可能地保留信号中的弱特征成分.利用LSD和PMD公式从理论和实践两方面证明该方法在有效抑制背景噪声的同时还可以降低音乐噪声,获取最大的信噪比,减少语音失真,实现比常规谱减法更好的语音增强效果.

Modified speech enhancement algorithm based on least mean square error estimation

On the basis of the theory of conventional amplitude subtraction and aimed at the problem concerned with noise elimination,a modified algorithm was presented,where the voiceless and voiced sound was processed with enhancement of high frequency and widening of frequency band,respectively,so that the features of weak components of the signal was reserved as possibly in the process of noise elimination.Meanwhile,LSD and PMD formulae were used to theoretically and practically verify that this method could restrain noise from background,reduce residual musical noise,gain maximum signal-to-noise ratio,and reduce speech distortion,realizing better enhancement effect than amplitude subtraction.