Phoneme based speaker verification by Gaussian mixture models with adaptation of subsets of dominant parameters and phonemes

This paper reports further results of speaker verification based on modelling the speaker´s phoneme by GMM (Gaussian mixture models) (D. Gutman et al, European SIGPRO Conf., Sept. 2002). In this approach, the speaker´s phonemes are represented by phoneme adapted Gaussian mixture models (PA-GMM) that are basically models for phonemes of speakers obtained by Bayesian adaptation of general phoneme-independent speaker models.