A discriminative performance metric for GMM-UBM speaker identification

Gaussian mixture modeling with universal background model (GMM-UBM) is a widely used method for speaker identification, where the GMM model is used to characterize a specific speaker’s voice. The estimation of model parameters is generally performed based on the maximum likelihood (ML) or maximum a posteriori (MAP) criteria. In this way, interspeaker information that discriminates between different speakers is not taken into account. To overcome this limitation, we design a discriminative performance metric to capture interspeaker variabilities leading to improve the classification capability of speaker models. A learning algorithm is presented to tune the Gaussian mixture weights by optimizing the frame classification accuracy of GMM classifiers. We design an objective function to directly relate the model parameters to the performance metric. The comparative study of the proposed method is done with the basic GMM-UBM system on the 2001 NIST SRE corpus. Experimental results demonstrate that the proposed learning algorithm considerably improves the GMM-UBM system on speaker identification.