In this paper we propose an algorithm for combined acoustic echo contr
ol and noise reduction. The algorithm is developed on the basis of a m
inimum mean-square error criterion and consists of a (conventional) ec
ho canceller and an additional noise and residual echo reduction filte
r. A special feature of the algorithm is the procedure for estimating
the residual echo power spectral density which relies on the assumptio
n that the phase of the estimated echo is approximately equal to the p
hase of the true echo. This assumption is verified by experimental res
ults. The residual echo power density estimate and the noise power den
sity estimate are then adaptively combined and used as an argument for
some spectral weighting rule such that the residual echo is attenuate
d and effectively masked by a low level of intentionally left backgrou
nd noise. The paper concludes with experimental results for a typical
car environment. (C) 1998 Elsevier Science B.V. All rights reserved.