一种基于变步长LMS算法的语音增强方法

LMS算法在自适应滤波器中得到广泛应用，但这种方法具有收敛速度慢，对非平稳环境敏感性强，步长需要谨慎选择才能达到收敛和失调的折中等缺点。为了改善非平稳条件下FIR自适应滤波器的性能，文章介绍了一种变步长的LMS算法，这种算法迭代过程中步长在规定的上下限内是关于信噪比的递减函数，用于自适应噪声对消器中去除含噪语音信号中的加性噪声，以解决固定LMS算法中跟踪速度和失调的矛盾。对不同信噪比的含噪语音信号去噪，仿真结果证明该方法优于NLMS(Normalized Least Mean Square)算法，在提高收敛速度的情况下减小了剩余均方误差和失调，但需增加少量的运算量。

A new speech enhancement method based on variable step size LMS algorithm

In this paper, a least mean squares(LMS) algorithm with adjustable step size is proposed for performance improvements of adaptive FIR filters in non-stationary environments. The step size varies between two hard limits based on a predetermined decreasing function of SNR estimated at each iteration step of the algorithm. Simulation of speech denoising shows that the proposed algorithm is superior to the normalized LMS(NLMS) algorithm in reducing the trade off between misadjustment and tracking ability, but requires considerably more computation.