Phonetically trained models for speaker recognition

In this paper, a speaker recognition system that introduces acoustic inform ation into a Gaussian mixture model (GMM)-based recognizer is presented. Th is is achieved by using a phonetic classifier during the training phase. Th e experimental results show that, while maintaining the recognition rate, t he decrease in the computational load is between 65% and 80% depending on t he number of mixtures of the models. (C) 2001 Acoustical Society of America .