In this paper the speaker identification system developed at Athens Information Technology is presented. It is based on the Gaussian Mixture modeling of the Mel-Frequency Cepstral Coefficients of speech. Starting from this basic algorithm, we describe and discuss two significant modifications that have resulted in performance enhancements, in terms of both processing speed and identification accuracy. We present the performance of our system in the recent CLEAR 2006 evaluation workshop and also discuss approaches to further improve our system by fusing decisions derived from a multitude of sensors in a multi-microphone setup.