A TENTATIVE ALGORITHM FOR ESTIMATING THE GLOTTAL WAVE-FORM WITH GLOBAL CLOSURE INFORMATION

Iris known that the glottal waveform includes acoustic information on the individuality of speakers, and that the inverse filtering method i s useful in extracting glottal waveform from speech signals. In conven tional methods, stationary analysis is performed for the duration of a few pitches. Short-time analysis with the drawback of unstable estima tion, is performed for one pitch duration. We propose a new method of inverse filtering by using the estimation algorithm for the vocal trac t transfer function; the analysis is performed stably for each pitch u sing the mapping algorithm with the Fejer kernel. The experimental res ults show reasonable estimation of glottal waveforms. In addition, we investigate the effect of the estimation order and the method of solvi ng the Yule-Walker equation. The experiments are performed for (1) dif ferent speakers, and different vowels in different order, and (2) diff erent algorithms such as the TLS and Levinson algorithm. (C) 1998 Scri pts Technica.