基于非因果先验信噪比估计的语音增强方法

Wiener Filter Speech Enhancement Based on Noncausal A Priori SNR Estimator

传统维纳滤波语音增强方法采用直接判决法来估计先验信噪比,直接判决法利用当前帧和当前帧之前的信息,这种方法对信号突变不能很好的估计,且在低信噪比区域的平滑程度也不够。针对上述两个缺点,提出一种基于非因果先验信噪比估计器的维纳滤波语音增强方法,这种方法利用当前帧之前、当前帧和当前帧之后的信息来联合估计先验信噪比。由于利用当前帧之后的信息,这种方法能够对信号突变进行很好的估计,且在低信噪比区域对后验信噪比进行深度平滑。实验结果表明,本文的方法优于传统的维纳滤波语音增强方法。

Conventional wiener filter speech enhancement method uses decision-directed method to estimate priori SNR. As decision-directed method only uses the data of the current frame and the data before current frame, it could not quickly respond to an abrupt increase in posteriori SNR, and also could not achieve an enough smoothness of posteriori SNR in the low SNR area. In order to avoid shortcomings of the decision-directed method, a new approach to the enhancement of noisy speech based a noncausal a priori SNR estimator is proposed. By using some data after current frame, the noncausal a priori SNR estimator could respond fast to an abrupt increase in posteriori SNR, and achieve an enough smoothness of posteriori SNR in the low SNR area. Experimental results indicate that the proposed approach outperforms the conventional method.