Eigenfeatures and Supervectors in Feature and Score Fusion for SVM Face and Speaker Verification

Eigenfaces are the classical features used in face recognition and have been commonly used with classification techniques based on Euclidean distance and, more recently, with Support Vector Machines. In speaker verification, GMM has been widely used for the recognition task. Lately, the combination of the GMM supervector, formed by the means of the Gaussians of the GMM, and SVM has resulted successful. In some works, dimensionality reduction transformations have been applied upon the GMM supervectors using Euclidean distance based classification methods to obtain eigenvoices. In this paper, eigenvoices will be used in a SVM system, and the fusion of eigenfaces and eigenvoices will be performed in a multimodal fusion. In addition to this, different feature and score normalization techniques will be applied before the classification process. The results show that the dimensionality reduction techniques do not improve the error rates provided by the GMM supervector and that the use of SVM and the multimodal fusion significantly increase the performance of the recognition systems.