Subspace-based speech enhancement with rank-deficient prewhitening

We study a new subspace-based noise reduction technique that can handle the case of narrowband noise, i.e. the case where the noise covariance matrix is rank-deficient. The formulation of the technique is based on the quotient singular value decomposition (QSVD) and is a generalization of existing techniques that can handle only broadband noise. We also show by examples that we are able to achieve a satisfactory noise reduction result.