Reverberant signal separation using optimized complex sparse nonnegative tensor deconvolution on spectral covariance matrix

Abstract

In this paper, an optimized complex nonnegative tensor factor 2D deconvolution (CNTF2D) is proposed to separate the sources that have been mixed in an underdetermined reverberant environment. Unlike conventional methods, the proposed model decomposition is performed directly on the statistics in the form of spectral covariance matrix instead of the data itself (i.e. the mixed signal). For faster convergence the model is adapted under the hybrid framework of the generalized expectation maximization and multiplicative update algorithms. This paper also proposes a solution to the issue of optimizing the model order i.e., number of components and convolutive parameters in the CNTF2D model. To this end, a latent-observation model based on Gamma-Exponential process is developed. In addition, the proposed Gamma-Exponential process can be used to initialize the parameterization of the CNTF2D. The proposed algorithm encodes a set of variable sparsity parameters derived from the Gibbs distribution. This permits a stable update and optimizes the CNTF2D with the correct degree of sparseness in the time-frequency domain. Experimental results on the underdetermined reverberant mixing environment have shown that the proposed algorithm is effective at separating the mixture with an average signal-to-distortion ratio of 2.5dB.

Multimode Process Monitoring Method Based on Multiblock Projection Nonnegative Matrix Factorization

“…At present, the initialization schemes often used are random value, singular value decomposition (SVD), and principal component analysis (PCA) [13,14,18]. Wang et al pointed out that using the PCA's load matrix as the initial value of W will make NMF converge faster and minimize errors [18].…”

“…After that, they proposed a mono audio separation method based on multicomponent nonnegative matrix factor two-dimensional deconvolution (NMF2D) [13]. Woo et al proposed an optimized complex nonnegative tensor factor two-dimensional deconvolution (CNTF2D) to separate sound sources mixed in an uncertain reverberation environment [14]. In recent decades, NMFbased methods have been used more and more widely.…”

Target Signal Extraction Method Based on Enhanced ICA with Reference

“…where μ and λ are also arbitrary positive scaling factors. e cost function in equation (17) and the cost function in equation (13) are reciprocal relations, while the cost function in equation 18is positive and negative with the cost function in equation (14). en, the corresponding EICA-R scheme is shown in the following expressions: Mathematical Problems in Engineering e scheme of (19) and (20) is corresponding to the scheme of (15) and (16).…”

“…It should be pointed out that ICA is not suitable for underdetermined cases. For underdetermined cases, we can adopt sparse analysis [11][12][13][14], deconvolution [15,16], and other methods. is paper only considers non-underdetermined cases.…”

Efficient Noisy Sound-Event Mixture Classification Using Adaptive-Sparse Complex-Valued Matrix Factorization and OvsO SVM

“…Approximate sparsity is an important consideration as they represent important information. Many sparse solutions have been proposed in the last decade [19][20][21][22][23][24][25]. Nonetheless, the optimal sparse solution remains an open issue.…”