Parametrical patterns of voices in forensic applications

Downloads

Authors

  • W. MAJEWSKI Institute of Telecommunications and Acoustics, Wrocław University of Technology, Poland
  • Cz. BASZTURA Institute of Telecommunications and Acoustics, Wrocław University of Technology, Poland

Abstract

A method of speaker recognition based on visual comparisons of sets of voice patterns is presented. These sets contain graphical patterns obtained from parameter extraction in the time-domain, frequency-domain and LPC (Linear Predictive Coding). In comparison to classical spectrograms, the parametrical patterns permit a better adaptation to specific speech signals that are available in forensic applications. Since parametrical patterns are more explicit than spectrograms, the decision reached by an expert may be more objective. The laboratory experiments on visual comparison of selected sets of patterns permitted to observe that in parametrical patterns of voices the intraspeaker variations are generally smaller than the interspeaker ones. The presented method of speaker recognition has been verified in a real-life case yielding valuable information on particular speakers. The results of laboratory experiments and real-life applications indicate that the visual method of speaker recognition based on various parametrical patterns of selected speech segments constitutes a very useful tool supporting the evidence proceeding with a much higher confidence level than it could be obtained on the basis of the classical spectrograms.

References

[1] R.H. BOLT et al., On the theory and practice of voice identification, National Academy of Sciences, Washington, D.C. 1979.

[2] H. HOLLIEN, The acoustics of crime - The new science of forensic phonetics, Plenum Press, New York 1990.

[3] W. MAJEWSKI, C. BASZTURA, Integrated approach to speaker recognition in forensic applications, Forensic Linguistics: Speech, Language and the Law, 3, 1, 50-64 (1996).