Comparison of glottal pulse shapes derived from linear prediction residual and pole-zero model residual of speech
B. Yegnanarayana · The Journal of the Acoustical Society of America · 1979
An important problem in voiced speech analysis is the determination of the characteristics of glottal pulses from speech signal. These characteristics can be derived in principle from the residual signal obtained by passing speech signal through the inverse of a model system for the vocal tract. The accuracy of the glottal pulse information thus derived depends on the accuracy of the model representing the vocal tract system. In this study glottal pulse shapes obtained by using an all-pole model and a pole-zero model are compared. The all-pole model is obtained from linear prediction analysis and the pole-zero model from a pole-zero decomposition technique. For most voiced sounds both the methods yield the same glottal pulse shapes. But for nasals and voiced fricatives the pole-zero model provides a more accurate characterization of glottal pulses than the all pole model. The results are explained on the basis of spectral approximation provided by the models.