Residual Echo Estimation with the Help of Minimum Statistics

Authors: M. Kallinger, J. Bitzer, K.-D. Kammeyer
Abstract: This contribution deals with a new technique to estimate the residual echo at the output of an acoustic echo canceller (AEC). It is known that the compensation filter of the AEC has to be very long in reverberant environments in order to provide sufficient echo attenuation. However, high filter orders involve long convergence periods during initialisation or after modifications of the echo path's impulse response. In these periods poor echo attenuation leads to undesired artifacts. One possible solution to this problem is to use a low-order and quickly converging AEC in combination with a Wiener post-filter after the AEC. Therefore, the power spectral density (PSD) of the residual echo has to be estimated properly. However, the problem is to achieve reliable estimates in periods of double-talk or during modifications of the echo path impulse response. This paper suggests a new method for estimating the residual echo with the help of minimum statistics to suppress interferences by near-end speech during double-talk situations.
Document type: Conference Paper
Publication: Leuven, Belgium, 21. - 22. March 2002
Conference: 3rd IEEE Benelux Signal Processing Symposium (SPS 02)
Index: 191
Files:
Paper
SPS_2002_kallinger.pdf83 KBSPS_2002_kallinger.ps.gz45 KB
BibTEX
Last change on 24.04.2008 by Admin
AIT ieee GOC tzi ith Fachbereich 1
© Department of Communications Engineering - University of BremenImprint / Contact