频域内的卡尔曼滤波语音增强算法

Speech Enhancement Algorithm Using Kalman Filter in Frequency Domain

提出了在频域内实现的卡尔曼滤波算法,该算法利用语音和噪声幅度谱的时变特性,先对语音幅度谱进行初步修正,提取较为准确的LPC系数,然后在每一频率点下对语音幅度用卡尔曼滤波进行递推估计,最终得到效果更好的增强语音。实验结果表明,本文算法有效地提高了增强语音的SNR,尤其是在高信噪比的情况下,效果更加明显。

A new Kalman filter method in the frequency domain which reeursively estimates the magnitude spectrum for each frequency using the time-varying characteristics of the speech and the noise is presented, and to get the more accurate LPC coefficients, the magnitude spectrum of the speech is preliminarily modified. Experimental results show that the proposed algorithm effectively improves SNR of the enhanced speech, especially in the case of high Signal-to-Noise Ratio（SNR）, the algorithm is much better than the time-domain Kalman filter.