Decision trees with improved efficiency for fast speaker verification

Classification and regression trees (CART) are convenient for low complexity speaker recognition on embedded devices. However, former attempts at using trees performed quite poorly compared to state of the art results with Gaussian Mixture Models (GMM). In this article, we introduce some solutions to improve the efficiency of the tree-based approach. First, we propose to use at the tree construction level different types of information from the GMM used in state of the art techniques. Then, we model the score function within each leaf of the tree by a linear score function. Considering a baseline state of the art system with an equal error rate (EER) of 8.6% on the NIST 2003 evaluation, a previous CART method provides typical EER ranging between 16% and 18% while the proposed improvements decrease the EER to 11.5%, with a computational cost suitable for embedded devices.