基于分数阶域最优滤波的语音增强算法

A Method for Speech Enhancement Based on fractional domain optimal filtering

本文介绍了一种基于分数阶域最优滤波的语音增强新算法。当存在畸变系统对信号造成畸变或者不能通过坐标轴旋转来完全解除信号和噪声的时频耦合时,本文将利用一种更具普适性的滤波方法,即采用最小均方误差准则下的分数阶最优滤波消除畸变和噪声的影响。仿真表明,在最小均方误差准则下求得的最优滤波算子并不一定是全局最小的,但是在不增加额外计算负担的情况下,能够获得比传统滤波更好的效果。

This paper introduces a new optimal filter speech enhancement based on fractional domain. When there is signal caused by distortion system or can not coordinate axis to completely lift the time-frequency of the signal and noise coupling, the paper will use a more universal filtering method, which uses the minimum mean square error criterion fractional optimal filter to eliminate distortion and noise. Simulation results show that under the minimum mean square error criterion obtained the optimal filtering operator is not necessarily a global minimum, but without additional computational burden of the situation, to obtain better results than the traditional filter.