组合FIR滤波器的语音增强算法研究

Study of Speech Enhancement Algorithm Based on the Combination of FIR Filters

在语音增强系统中,当一个语音信号相对于背景噪声很强时,采用传统的LMS算法存在剩余均方误差随有用信号功率线性增加的问题,因此提出B-WSLMS算法。与传统的语音增强算法不同,B-WSLMS算法将WSLMS算法应用到组合FIR滤波器中,算法的权值根据输出功率变化进行更新,并且权值更新收敛系数与输出功率成反比进行变化,有效弥补了LMS算法的缺点。同时,通过在组合FIR滤波器中选取不同的辅助变量收敛系数,B-WSLMS算法可以使语音增强系统得到更好的信噪比提高量。仿真结果表明,在语音增强系统中,B-WSLMS算法优于LMS算法及WSLMS算法.

In speech enhancement systems,if the speech signal is stronger than the noise signal,there is a problem that the filter weight vector increases with the available signal power linearly using the tradition LMS algorithm,so the B-WSLMS algorithm is proposed.The B-WSLMS algorithm applies the WSLMS algorithm to the combination of FIR filters,and because the step size of the adaptive the filter weight vector can update by the output power of the system,which can make up the weakness of the LMS algorithm.By simulation results,the B-WSLMS algorithm has a better performance in speech enhancement in a sense of mean-squares-error and speech quality improvement than the LMS and the WSLMS algorithms.