Mutual Information Minimization for Under-Determined Blind Source Separation

Author

Wang, Fuxiang ; Zhang, Jun

Author_Institution

Sch. of Electron. & Inf. Eng., Beihang Univ., Beijing, China

fYear

2009

fDate

11-13 Dec. 2009

Firstpage

1

Lastpage

4

Abstract

An important step of sparse representation technique for under-determined BSS (blind source separation) is the estimation of the mixing matrix. In this paper, a new method to estimate the mixing matrix is proposed. The objective is to find the mixing matrix to minimize the mutual information of the estimated sources. An algorithm for the learning of the mixing matrix is proposed by the natural gradient. Simulation results of speech separation demonstrate the effectiveness o f our method.

Keywords

blind source separation; learning (artificial intelligence); matrix algebra; signal representation; speech processing; learning algorithm; mixing matrix estimation; mutual information minimization; natural gradient method; sparse representation technique; speech separation; under-determined blind source separation; Blind source separation; Clustering algorithms; Entropy; Laplace equations; Mutual information; Probability density function; Source separation; Sparse matrices; Speech;

fLanguage

English

Publisher

ieee

Conference_Titel

Computational Intelligence and Software Engineering, 2009. CiSE 2009. International Conference on

Conference_Location

Wuhan

Print_ISBN

978-1-4244-4507-3

Electronic_ISBN

978-1-4244-4507-3

Type

conf

DOI

10.1109/CISE.2009.5365757

Filename

5365757