Speaker verification using time cepstral principal components derived from a pole-zero model

A speaker verification approach is presented that integrates pole-zero cepstral coefficients with a speech parameterization technique that utilizes time cepstral principal components (TCPC´s). TCPC analysis utilizes static and dynamic speaker characteristics, and the pole-zero cepstral coefficients emphasize the formants. Results are compared for the same approach using cepstral coefficients derived from higher order LP analysis (20th order) using a Gaussian mixture model for verification. Results indicate that the TCPC approach using pole-zero cepstral coefficients improves the verification accuracy over the same approach using cepstral coefficients derived from higher order LP analysis.