SPEAKER DISCRIMINATION USING WIENER CANONICAL FORMS.
Technical rept. no. 18,
INFORMATION RESEARCH ASSOCIATES INC WALTHAM MASS
Pagination or Media Count:
Presented in this report are the results of one experiment conducted in our overall effort aimed at determining the applicability of the Wiener Canonical Expansion WCE in distinguishing classes of real stochastic waveforms. The two classes of waveforms selected were the time speech waveforms from two different speakers. No special consideration was given to the fact that these sample waveforms represented speech. Three separate experiments consisting of voiced sounds, unvoiced sounds and regular speech were conducted. The tentative result of this investigation is that over the range of Laguerre time scaling coefficients used, speaker identification cannot be accomplished with any degree of certainty by treating a speaker waveform as a sample from a stationary or quasi-stationary random process. The necessary theoretical background, the experimental setup and the results on which this conclusion is based form the body of this report. Author
- Voice Communications