Bilinear Time Series Model as an Alternative Way of Speaker Modeling

  • Oskar Kochana
  • Patrycja Ksiazek
  • Michal Olszak
  • Ewa Bielinska
Conference paper
Part of the Advances in Intelligent Systems and Computing book series (AISC, volume 289)


In the paper a class of non-linear time series models is considered, with respect to possible application for speaker recognition. Registered speech signal is is a non-stationary time series. This non-stationarity is usually modeled as autoregressive time series with time varying parameters. In the paper a bilinear approximation of non-stationary autoregressive model is proposed. This way, a model with time varying parameters is approximated by a constant parameters model. Parameters of the bilinear model are assumed to be the speaker features,and are applied for speaker recognition. Effectiveness of the proposed method is compared with classic methods of speaker recognition.


non-stationary AR models bilinear time series model speech analysis speaker recognition recognition 


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.


  1. 1.
    Atal, B.S.: Effectiveness of linear prediction characteristics of the speech wave for automatic speaker identification and verification. J. Acoust. Soc. Am. 55, 1304–1312 (1974)CrossRefGoogle Scholar
  2. 2.
    Furui, S.: Recent advances in speaker recognition. Pattern Recognition Letters 18(9), 859–872 (1997)CrossRefGoogle Scholar
  3. 3.
    Bielinska, E.M.: Bilinear representation of non-stationary autoregressive time series. In: Proceedings of the International Conference on System Science 2013. Advances in Systems Science, pp. 737–746. Springer, Heidelberg (2013)Google Scholar
  4. 4.
    Bensty, J., Sondhi, M., Huang, Y. (eds.): Springer Handbook of Speech Processing. Springer, Heidelberg (2007)Google Scholar
  5. 5.
    Campbell, J.: Speaker recognition: A Tutorial. Proceeding of the IEEE 85(9) (1977)Google Scholar
  6. 6.
    Durbin, J.: Effcient estimation of parameters in moving-average models. Biometrika 46, 306–316 (1959)CrossRefMATHMathSciNetGoogle Scholar
  7. 7.
    Kohlmorgen, J., Lemm, S.: An On-lLine method for segmentation and identification of non-stationary time series. In: Proceeding of Neural Networks for Signal Processing XI, pp. 113–122 (2001)Google Scholar
  8. 8.
    Ludwig, M.: Building on Durbin’s method to estimate MA processe. Improving Durbin’s method to estimate MA processes, arXiv:1304.7956,
  9. 9.
    Ni, H., Yin, H.: Self-organising mixture autoregressive model for non-stationary time series modelling. Int. Journal of Neural Systems. Dec. 18(6), 469–480 (2008)CrossRefGoogle Scholar
  10. 10.
    Ozaki, T., Tong, H.: On moving average parameter estimation. In: Proceedings of the 8th Hawaii International Conference on System Science, pp. 224–226 (1995)Google Scholar
  11. 11.
    Pollock, D.S.G.: A Handbook of time series analysis, signal processing and dynamics. Academic Press (1999)Google Scholar
  12. 12.
    Rabiner, L.R.: Fundamentals of Speech Recognitions. Prentice Hall, New Jersey (1993)Google Scholar
  13. 13.
    Sandgren, N., Stoica, P., Babu, P.: On moving average parameter estimation. In: Proceedings of the 20th European Signal Processing Conference (EUSIPCO), pp. 2348–2351 (2012)Google Scholar

Copyright information

© Springer International Publishing Switzerland 2014

Authors and Affiliations

  • Oskar Kochana
    • 1
  • Patrycja Ksiazek
    • 1
  • Michal Olszak
    • 1
  • Ewa Bielinska
    • 1
  1. 1.The Silesian Technical UniversityGliwicePoland

Personalised recommendations