Section: New Results

Discriminative learning for Automatic speaker recognition

Participants : Reda Jourani [correspondant] , Khalid Daoudi, Régine André-Obrecht, Driss Aboutajdine.

We continued our work aiming at developing efficient versions of Large Margin Gaussian Mixture Models (LM-GMM) for speaker identification. We developed a new and efficient learning algorithm and evaluated it on NIST-SRE'2006 data. The results show that, combined with the channel compesentation technique SFA, this new algorithm outperforms the state-of-the-art discriminative method GMM-supervectors SVM combined with NAP compensatation.

  • Related publication: [10] .