Analysis of broadband GEVD-based blind source separation

Tools

Redif, Soydan and Pestana, Jennifer and Proudler, Ian K.; (2019) Analysis of broadband GEVD-based blind source separation. In: ICASSP 2019 - 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, Piscataway, NJ, pp. 8028-8032. ISBN 9781479981311 (https://doi.org/10.1109/ICASSP.2019.8683237)

[thumbnail of Redif-etal-ICASSP2019-Analysis-of-broadband-GEVD-based-blind-source-separation]
Preview
 Text. Filename: Redif_etal_ICASSP2019_Analysis_of_broadband_GEVD_based_blind_source_separation.pdf
Accepted Author Manuscript
Download (470kB)| Preview

Abstract

One approach to blind source separation of instantaneously mixed, non-stationary sources involves using the generalized eigenvalue decomposition of two estimated covariance matrices. The assumption is that the source statistics change with time whilst the mixing matrix does not. A recent generalisation of this approach to convolutive mixtures was achieved by extending the generalized eigenvalue decomposition to polynomial matrices. In this paper, we present a further investigation into this broadband BSS technique. We derive some expressions for the conditions under which source separation is possible. The validity of our analysis is illustrated through some computer simulations.

ORCID iDs

Redif, Soydan, Pestana, Jennifer ORCID logoORCID: https://orcid.org/0000-0003-1527-3178 and Proudler, Ian K.;
CORE (COnnecting REpositories)