DISCRIMINANT APPROACHES FOR GMM BASED SPEAKER DETECTION SYSTEMS
Résumé
This paper presents some experiments on discriminative training for GMM/UBM based speaker recognition systems. We propose two MMIE adaptation methods for GMM component weights suitable for speaker recognition. The impact on performance of this training methods is compared to the standard weight estimation/adaptation criterion, MLE and MAP on standard GMM based systems and on SVM based systems. The results enforce the difficulty to introduce discriminative behaviour in a GMM based system whereas it is inherent in SVM based systems.