Output details

Speech Signal Enhancement based on MAP Algorithm in the ICA Space

<22> This paper introduces a new approach to enhancement of speech signals corrupted by additive noise. State-of-the-art speech enhancement techniques perform the enhancement in a transform domain. Unlike conventional methods that employ a single fixed transformation, the proposed method estimates a separate transformation from signal and noise and employs these during the enhancement process. This enables to capture better the statistical properties of the signal and noise. It is the first time such use of multiple transformations was presented. This work was in collaboration with Shandong University (China) and led to £356k EPSRC funding.

Note: It has led to invitation for contribution to an edited book below

P. Jancovic, X. Zou, M. Köküer, (2011) “Speech Enhancement and Representation Employing the Independent Component Analysis,” in: Recent Advances in Robust Speech Recognition Technology, Eds. J. Ramirez and J. Gorriz, Bentham Science Publishers, Feb. 2011, pp. 103-113.