Robust speaker verification with optimal pitch bases expansions

  • Authors:

  • Robert M. Nickel

  • Affiliations:

  • The Pennsylvania State University, University Park, PA

  • Venue:
  • WISICT '04 Proceedings of the winter international synposium on Information and communication technologies
  • Year:
  • 2004

Quantified Score

Hi-index 0.00

Visualization

Abstract

We are presenting a new method to improve the robustness of speaker verification algorithms. It is based on an innovative approach that departs radically from commonly used mechanisms for speech recognition and speaker verification. In addition to speech features derived from spectral analysis, we are using speech features derived from orthogonal expansions of pitch period segments. The expansions are designed to optimally match the characteristics of a given speaker. Our research demonstrates that the presented strategy delivers a performance that is superior to traditional methods in the context of text-dependent speaker verification with cooperative speakers in a low noise environment. In experiments we were able to reduce the probability of impostor access by more than 50%. The overall complexity of the proposed algorithm is comparable to the complexity of algorithms in current commercial products.