ROBUST AND ELEGANT, PURELY STATISTICAL ADAPTATION OF ACOUSTIC ECHO CANCELER AND POSTFILTER

ABSTRACTWe presentanextremelysimpleandeffectivesignalpro-cessing solution to the acoustic echo control problem. Theapproach is based on the concept of synchronous statisticaladaptation of an acoustic echo canceler and a postfilter forresidual echo suppression. The required convergence stateof the echo canceler is estimated by a new statistical ele-ment of our system. Double talk detection is not requiredexplicitly. The whole solution combines an excellent outputsignal quality with a high degree of elegance and simplicity.1. INTRODUCTION AND SYSTEM OVERVIEWFast and robust adaptation of the acoustic echo control(AEC) unit in hands-free communication systems is oftenconsidered to be difficult in non-stationary noisy environ-ments. The adaptation of many AEC units explicitly relieson the sophisticated detection and classification of differ-ent acoustic events (e.g. room impulse response changes ordouble talk situations). In our paper we will demonstratethat the AEC problem can be solved in a simpler, purelystatistical framework, at least if the AEC unit relies on thefilter arrangement with acoustic echo canceler and postfilterfor residual echo suppression as shown in Figure 1.