Study of Speech Enhancement System Using Combinational Adaptive Filtering

When using the tradition LMS algorithm in speech enhancement systems, there is a problem of the main-square-error increasing with the available signal power linearly, so the WSLMS algorithm which updates the step size of the adaptive filter weight vectors by the output power of the system is presented at first. In order to improve adaptive filter performance, then the CC-WSLMS algorithm is proposed using the WSLMS algorithm as component filter in the convex combination of two adaptive filters. By simulation results, it is illustrated that a better performance of speech enhancement systems in a sense of mean-squares-error and a better speech quality improvement are achieved via the CC-WSLMS algorithm than the LMS and the WSLMS algorithm.