ETRW on Speech Processing in Adverse Conditions Cannes-Mandelieu, France November 10-13, 1992

Speech Enhancement by Adaptive Wiener Filtering based on Cumulant AR Modelling

E. Masgrau (1), J. Salavedra (2), Asunción Moreno (2), A. Ardanuy (2)

(1) Department of Electrical Engineering and Computers, Centro Politecnico Superior de Ingenieros, University of Zaragoza, Spain
(2) Department of Signal Theory and Communications, E.T.S.I. de Telecomunicacion, Politechnic University of Catalonia, Barcelona, Spain

In this paper we study some speech enhancement algorithms based on the iterative Wiener method due to Lim and Oppenheim [1], but where the AR spectral estimation of the speech is carried out using a third order cumulant analysis. This work extends some preceding papers due to the authors, providing a detailed analytical study of the convergence of the iterative algorithms. This analysis allows to understand the behaviour of the original [1] and the proposed cumulant algorithms, providing some insigths and suggestions to develop new and improved algorithms. An exhaustive empirical analysis establishs that the hybrid algorithm presented in [3] by the authors outperforms the original correlation algorithm, specially at low SNR largerly.

Full Paper

Bibliographic reference.  Masgrau, E. / Salavedra, J. / Moreno, Asunción / Ardanuy, A. (1992): "Speech enhancement by adaptive Wiener filtering based on cumulant AR modelling", In SPAC-1992, 143-146.