Speech Enhancement Using Kernel Recursive Least-Squares Method

In this paper, we propose a new speech enhancement structure based on kernel recursive least squares adaptive filtering. The combination of the famed kernel trick andrecursive least squares (RLS) algorithm yields powerful nonlinear extensions, named collectively here as KRLS. This method improves the adaptive filtering performance innonlinear adaptive filtering scenarios. We compare the performance of this kernel based algorithm in the area of dualchannelspeech enhancement with other linear adaptive filtering techniques. Experimental results show that the proposed enhancement structure has better performance in a sense of mean-squares error (MSE) and speech quality improvement than the those based on standard LMS, Normalized LMS, Affine projection, and conventional RLS algorithms

Keywords:

Speech Enhancement , Linear Adaptive Filtering , Reproducing Kernel Hilbert Spaces , KernelMethods , Kernel Adaptive Filtering