Subspace-based speech enhancement and implementation

Abstract

Signal subspace approach for speech enhancement is based on linear algebra and matrix analysis, the main idea is to decompose the corrupted speech into two subspaces, which one of them is the noise subspace. By nulling the noise subspace one can achieve the estimated clean speech. In the past years, much research has been focused on the theoretical aspects; however, not so much attention was paid to issues concerning a practical implementation of a noise reduction system. Hence, a detailed practical implementation will be provided, as result shows that with improved techniques it can be applied in real life communication applications.