Date of Award

3-1996

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Department of Electrical and Computer Engineering

First Advisor

Dennis W. Ruck, PhD

Abstract

Classification of time series has wide Air Force, DoD and commercial interest, from automatic target recognition systems on munitions to recognition of speakers in diverse environments. The ability to effectively model the temporal information contained in a sequence is of paramount importance. Toward this goal, this research develops theoretical extensions to a class of stochastic models and demonstrates their effectiveness on the problem of text-independent (language constrained) speaker recognition. Specifically within the hidden Markov model architecture, additional constraints are implemented which better incorporate observation correlations and context, where standard approaches fail. Two methods of modeling correlations are developed, and their mathematical properties of convergence and reestimation are analyzed. These differ in modeling correlation present in the time samples and those present in the processed features, such as Mel frequency cepstral coefficients. The system models speaker dependent phonemes, making use of word dictionary grammars, and recognition is based on normalized log-likelihood Viterbi decoding. Both closed set identification and speaker verification using cohorts are performed on the YOHO database. YOHO is the only large scale, multiple-session, high-quality speech database for speaker authentication and contains over one hundred speakers stating combination locks. Equal error rates of 0.21% for males and 0.31% for females are demonstrated. A critical error analysis using a hypothesis test formulation provides the maximum number of errors observable while still meeting the goal error rates of 1% False Reject and 0.1% False Accept. Our system achieves this goal.

AFIT Designator

AFIT-DS-ENG-96-01

DTIC Accession Number

ADA322863

Share

COinS