USEFULNESS OF THE LPC-RESIDUE IN TEXT-INDEPENDENT SPEAKER VERIFICATION

This paper is a contribution to automatic speaker recognition. It cons iders speech analysis by linear prediction and investigates the recogn ition contribution of its two main resulting components, namely the sy nthesis filter on one hand and the residue on the other hand. This inv estigation is motivated by the orthogonality property and the physiolo gical significance of these two components, which suggest the possibil ity of an improvement over current speaker recognition approaches base d on nothing but the usual synthesis filter features. Specifically, we propose a new representation of the residue and we analyse its corres ponding recognition performance by issuing experiments in the context of text-independent speaker verification. Experiments involving both k nown and new methods allow us to compare the recognition performance o f the two components. First we consider separate methods, then we comb ine them. Each method is tested on the same database and according to the same methodology, with strictly disjoint training and test data se ts. The results show the usefulness of the residue when used alone, ev en if it proves to be less efficient than the synthesis filter. Howeve r, when both are combined, the residue shows its true relevance. It ac hieves a reduction of the error rate which, in our case, went down fro m 5.7% to 4.0%.