Gaussian selection applied to text-independent speaker verification

Fast speaker verification systems can be realised by reducing the computation associated with searching of mixture components within the statistical model such as a Gaussian mixture model, GMM. Several improvements regarding computational efficiency have already been proposed for speaker verification. In this paper, the technique of Gaussian selection is applied to the speaker verification task. Gaussian selection is commonly known in speech recognition where it is used to speed up the scoring process of HMM recognisers. Here we use the same technique to reduce the computation of mixture components within the GMM framework. Experiments compare different selection methods on the text-independent Odyssey 2001 speaker verification database. Further, the selection methods are compared with the baseline approach of scoring all mixture components in the full model. The results reveal a computational reduction of factor ten with only minor degradation in verification performance.