Speech blind signal separation with FastICA and Markov Chain combination

Author

Ziming Liu ; Zhenjiang Miao ; Lili Wan

Author_Institution

Inst. of Inf. Sci., Beijing Jiaotong Univ., Beijing, China

fYear

2014

fDate

19-23 Oct. 2014

Firstpage

541

Lastpage

544

Abstract

Applying Independent component analysis (ICA) on higher order statistics signal, can separate both the statistical independent and non-Gaussian source signals from the mixed signal. In many variants of ICA, fixed-point algorithm (also be known as FastlCA) is widely used in signal processing due to its fast convergence and good separation. Although the FastlCA achieves some promising separation results, but when the source signals are the non-sparse signals, the effect of the separation is not very ideal. Therefore, in this paper, the Bayesian estimation method based on Markov Chain Monte Carlo is proposed for improving these limitations.

Keywords

Gaussian processes; Markov processes; Monte Carlo methods; audio signal processing; blind source separation; fixed point arithmetic; higher order statistics; independent component analysis; Bayesian estimation method; Markov chain; Monte Carlo method; fastICA; fixed-point algorithm; higher order statistics signal; independent component analysis; mixed signal; nonGaussian source signals; nonsparse signals; source signal processing; speech blind signal separation; statistical independent signal; Abstracts; Algorithm design and analysis; Educational institutions; Markov processes; Speech; Bayesian methods; Independent component analysis; Markov chain; Monte Carlo methods; speech blind signal separation;

fLanguage

English

Publisher

ieee

Conference_Titel

Signal Processing (ICSP), 2014 12th International Conference on

Conference_Location

Hangzhou

ISSN

2164-5221

Print_ISBN

978-1-4799-2188-1

Type

conf

DOI

10.1109/ICOSP.2014.7015063

Filename

7015063