Cluster-Dependent Feature Transformation for Telephone-Based Speaker Verification

  • Chi-Leung Tsang
  • Man-Wai Mak
  • Sun-Yuan Kung
Conference paper
Part of the Lecture Notes in Computer Science book series (LNCS, volume 2688)


This paper presents a cluster-based feature transformation technique for telephone-based speaker verification when labels of the handset types are not available during the training phase. The technique combines a cluster selector with cluster-dependent feature transformations to reduce the acoustic mismatches among different handsets. Specifically, a GMM-based cluster selector is trained to identify the cluster that best represents the handset used by a claimant. Handset distorted features are then transformed by cluster-specific feature transformation to remove the acoustic distortion before being presented to the clean speaker models. Experimental results show that cluster-dependent feature transformation with number of clusters larger than the actual number of handsets can achieve a performance level very close to that achievable by the handset-based transformation approaches.


Gaussian Mixture Model Transformation Parameter Equal Error Rate Feature Transformation Speaker Model 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.


  1. [1]
    M.W. Mak and S.Y. Kung, “Combining stochastic feautre transformation and handset identification for telephone-based speaker verification,” in Proc. ICASSP’2002, 2002, pp. I701–I704.Google Scholar
  2. [2]
    A. Sankar and C.H. Lee, “A maximum-likelihood approach to stochastic matching for robust speech recognition,” IEEE Trans. on Speech and Audio Processing, vol. 4, no. 3, pp. 190–202, 1996.CrossRefGoogle Scholar
  3. [3]
    C. L. Tsang, M.W. Mak, and S. Y. Kung, “Divergence-based out-of-class rejection for telephone handset identification,” in Proc. ICSLP’02, 2002, pp. 2329–2332.Google Scholar
  4. [4]
    A.P. Dempster, N. M. Laird, and D.B. Rubin, “Maximum likelihood from incomplete data via the EM algorithm,” J. of Royal Statistical Soc., Ser. B., vol. 39, no. 1, pp. 1–38, 1977.zbMATHMathSciNetGoogle Scholar
  5. [5]
    D.A. Reynolds, “HTIMIT and LLHDB: speech corpora for the study of handset transducer effects,” in ICASSP’97, 1997, vol. 2, pp. 1535–1538.Google Scholar
  6. [6]
    Eric W.M. Yu, M. W. Mak, and S.Y. Kung, “Speaker verification from coded telephone speech using stochastic feature transformation and handset identification,” in Pacific-Rim Conference on Multimedia 2002, 2002, pp. 598–606.Google Scholar
  7. [7]
    S.B. Davis and P. Mermelstein, “Comparison of parametric representations for monosyllabic word recognition in continuously spoken sentences,” IEEE Trans. on ASSP, vol. 28, no. 4, pp. 357–366, August 1980.CrossRefGoogle Scholar
  8. [8]
    H.C. Wang C. S. Liu and C.H. Lee, “Speaker verification using normalized loglikelihood score,” IEEE Trans on Speech and Audio Processing, vol. 4, no. 1, pp. 56–60, 1996.CrossRefGoogle Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 2003

Authors and Affiliations

  • Chi-Leung Tsang
    • 1
  • Man-Wai Mak
    • 1
  • Sun-Yuan Kung
    • 2
  1. 1.Center for Multimedia Signal Processing Dept. of Electronic and Information EngineeringThe Hong Kong Polytechnic UniversityHong Kong SAR, China
  2. 2.Dept. of Electrical EngineeringPrinceton UniversityUSA

Personalised recommendations