Training Feature Xm} Speaker Speaker identification (SI) systems based on Gaussian MixSignal Extraction Modeling Speaker, s ture Models (GMMs) have demonstrated high levels of accuracy when both training and testing signals are acquired in (a) near ideal conditions. These same systems when trained and {Xg.. s } tested with signals acquired under non-ideal channels such as telephone have been shown to have markedly lower accuracy Test Signal Fte X Litklo Spae levels. In this paper, we consider a reverberant test environUnknown 3etraci n CopuaihoSdenai-ye ment and its impact on SI. We measure the degradation in SI Speaker, s v accuracy when the system is trained with clean signals but (b) tested with reverberant signals. Next, we propose a method whereby training signals are first filtered with a family of Fig. 1. (a) Training and (b) testing stages in SI reverberation filters prior to construction of speaker models; the reverberation filters are designed to approximate expected test room reverberation. Reverberant test signals are then scored against the family of speaker models and identification is made. Our research demonstrates that by approximatthe speaker's training feature vectors. The GMM-based aping test room reverberation in the training signals, the channel proach has shown to be very successful in accurately idenmismatch problem can be reduced and SI accuracy increased. tifying speakers from a large population [2]. In utilizing a GMM, we assume the probability density function (pdf) for feature vector x given speaker model A, can be modeled as a