Sixth European Conference on Speech Communication and Technology
(EUROSPEECH'99)

Budapest, Hungary
September 5-9, 1999

Modelling Output Probability Distributions for Enhancing Speaker Recognition

Jason Pelecanos, Sridha Sridharan

Speech Research Lab, RCSAVT, Queensland University of Technology, Brisbane, Australia

This paper discusses the use of a secondary likeli-hood classifier scheme for improving speaker recognition performance. The system models the out-put likelihoods of a typical Gaussian Mixture Model system across multiple speakers. The Out-put Probability Distributions (OPD) of the primary classifiers contain information on inter-speaker relationships, and are modelled by secondary classifiers to improve recognition accuracies. A com-parison of the OPD system with the traditional likelihood ratio and maximum likelihood scoring schemes for verification and identification is performed. Fusion of traditional measures with OPDs is shown to enhance overall recognition performance.


Full Paper (PDF)   Gnu-Zipped Postscript

Bibliographic reference.  Pelecanos, Jason / Sridharan, Sridha (1999): "Modelling output probability distributions for enhancing speaker recognition", In EUROSPEECH'99, 999-1002.