VARIANTS OF CHPSTKUM BASED SPEAKER IDENTITY VERIP
George A. Velius · 1988
One of the most promising techniques for Speaker Identity Verification (SIV) is a template matching scheme using cepstral analysis coefficients as the identity-bearing features of speech signals. This paper investigates analysis parameters and various distaice measures for this approach to SIV. Two parameters are matically varied the length of the signal analysis window, the order of the LPC-cepstrum analysis. Computational ssociated with the choice of parameters are also ed The measures tested are: the Euclidean, variance , differential mean weighting, Kahn's ed weighting, the Mahalanobis distance, and the Fisher discriminant. Using the Equal Error Rate (EER) of pairwise utterance dissimilarity distributions, performance is estimated for pre-specified and (a simulation of) user-determined input vocabulary. Performance vaies significantly across vocabulary, and average performance is approximately 5% EER for the better algorithms on telephone speech.