基于自适应共轭梯度参数估计的顽健卡尔曼滤波语音增强算法

对于加性噪声影响下的语音信号，利用双通道输入建立起来的增广卡尔曼滤波器模型，采用自适应共轭梯度方法对纯净语音和有色噪声干扰模型分别进行参数估计，提出了一种有效的语音增强算法。由于该方法对模型参数的估计精确性较高，而且估计速度快，同卡尔曼滤波类的其它语音增强方法相比，其语音增强效果良好，且具有一定的顽健性。仿真实验表明在环境噪声很复杂的情况下，该方法仍然有效。

Speech enhancement with Kalman-filters based on adaptive conjugate gradient parameter estimation

Using extended Kalman-filter model constructed through two channel inputs and conjugate gradient parameter estimation technique for both speech and noise models, an efficient speech enhancement algorithm is proposed for the speech signal with additive noise. Because this algorithm possesses higher parameter estimation accuracy and speed, it has better speech enhancement performance and robustness than that of other Kalman speech enhancement algorithms. Simulation results show that the proposed method is still valid under very complicated noise environment.