Adaptive Blind Signal and Image Processing: Learning Algorithms and Applications
Adaptive Blind Signal and Image Processing: Learning Algorithms and Applications
Underdetermined blind separation of speech signals with delays in different time-frequency domains
Nonlinear Speech Modeling and Applications
Source separation in post-nonlinear mixtures
IEEE Transactions on Signal Processing
| Hi-index | 0.01 |
The principal aim of this work is to demonstrate, from an empirical point of view, the effectiveness of a previously proposed technique for the Blind Source Separation (BSS) with Post Non Linear (PNL) underdetermined instantaneous mixing model (uBSS), in the more complex and realistic case where delayed sources are considered in the linear part of the mixing (PNL-uBSS with delays). The proposed approach is composed of two consecutive stages: in the first stage the inverse nonlinearities are estimated by Gaussianization of the mixtures; in the second stage source signals are extracted from the linearized mixtures using a three step approach already known in the literature for linear delayed uBSS. An improved technique based on Extended Gaussianization is also provided for the estimation of inverse nonlinearities. Experimental results using synthetic mixtures of real world data (speech signals) are given to prove the effectiveness of the proposed approach.