Kalman滤波语音增强算法的参数估计与实现

Parameter Estimation of Kalman Filter Speech Enhancement Algorithm and Its Realization

针对Kalman滤波语音增强算法的参数估计及实现问题,提出基于协作式的新型递归神经网络方法。该方法通过协作式递归神经网络算法进行自回归AR参数估计,并利用有噪自回归信号的最小均方算法进行噪声方差参数估计。仿真实验表明,综合以上2种参数估计方法的Kalman滤波语音增强算法具有高效、实现简单的特点,消噪效果明显。

Based on Cooperative recurrent neural network,this paper proposes a novel thought to solve the parameter estimation and the realization problem of Kalman filter speech enhancement algorithm . Through cooperative recurrent neural network for parameter estimation of autoregressive AR, and using LMS algorithm of the autoregressive signals in noise for parameter estimation of noise variance. The simulation results show that a combination of the above two methods in parameter estimation of Kalman filter speech enhancement algorithm has high efficiency, simple features, and the effect of noise elimination is obvious.