A Signal Subspace Speech Enhancement Approach Based On Low-rank And Sparse Matrix Decomposition

Subspace-based methods have been effectively used to estimate enhanced speech from noisy speech samples. In the traditional subspace approaches, a critical step is splitting of two invariant subspaces associated with signal and noise via subspace decomposition, which often is performed by singular-value decomposition or eigenvalue decomposition. However, these decomposition algorithms are highly sensitive to the presence of large corruptions, resulting in a large amount of residual noise within enhanced speech in low signal-to-noise ratio(SNR) situations. In this paper, a joint low-rank and sparse matrix decomposition(JLSMD) based subspace method is proposed for speech enhancement. In the proposed method, we firstly structure the corrupted data as a Toeplitz matrix and estimate its effective rank value for the underlying clean speech matrix. Then the subspace decomposition is performed by means of JLSMD, where the decomposed low-rank part corresponds to enhanced speech and sparse part corresponds to noise signal, respectively. An extensive set of experiments have been carried out for both of white Gaussian noise and real-world noise. Experimental results show the proposed method performs better than conventional methods in many types of strong noise conditions, in terms of yielding less residual noise and lower speech distortion. The innovations are described as follow:1. In the case of white noise, we proposed subspace speech enhancement algorithm based on JLSMD, which removed more residual noise and made the speech distortion degree lower than the traditional subspace algorithm2. In the case of colored noise, improved subspace speech enhancement algorithm based on JLSMD made its wide practicality, at the same time is better than the traditional speech enhancement algorithm based on SVD subspace.