Multiple background models for speaker verificatio

Authors: A K Sarkar S Umesh

Publish Date: 2012/06/15

Volume: 15, Issue: 3, Pages: 351-364

PDF Link

Abstract

In this paper we investigate the use of Multiple Background Models MBMs in Speaker Verification SV We cluster the speakers using either their Vocal Tract Lengths VTLs or by using their speaker specific Maximum Likelihood Linear Regression MLLR supervector and build a separate Background Model BM for each such cluster We show that the use of MBMs provide improved performance when compared to the use of a single/gender wise Universal Background Model UBM While the computational complexity during test remains same for both MBMs and UBM MBMs require switching of models depending on the claimant and also scorenormalization becomes difficult To overcome these problems we propose a novel method which aggregates the information from Multiple Background Models into a single gender independent UBM and is inspired by conventional Feature Mapping FM technique We show that using this approach we get improvement over the conventional UBM method and yet this approach also permits easy use of scorenormalization techniques The proposed method provides relative improvement in EqualError Rate EER by 1365  in the case of VTL clustering and 1543  in the case of MLLR supervector when compared to the conventional single UBM system When ATnorm scorenormalization is used then the proposed method provided a relative improvement in EER of 2096  for VTL clustering and 2248  for MLLR supervector based clustering Furthermore the proposed method is compared with the gender dependent speaker verification system using Gaussian Mixture ModelSupport Vector Machines GMMSVM supervector linear kernel The experimental results show that the proposed method perform better than gender dependent speaker verification system

Keywords: