High speed estimation of vocal area function using digital signal processor.
Hiroyuki Kamata, Yoshihisa Ishida, Yasuo Ogawa · IEEJ Transactions on Industry Applications · 1990
This paper describes the method to quickly estimate the vocal area function by using a digital signal processor (DSP). The method of estimation on the basis of a reflection coefficient derived by Levinson-Durbin algorithm is the best one. However, it is necessary to remove (i) glottis characteristic and (ii) radiation characteristic from the voice wave that is observed by a microphone, to estimate precisely the vocal area function that corresponds to a vocal of human.We paid attention to an adaptive inverse filter in this pre-processing. However, a conventional adaptive inverse filter is not suitable to DSP in terms of process and accuracy.This paper proposes the method that the effect of an adaptive inverse filter is given to autocorrelation directly. The proposed method reduces a calculation volume to 70% in comparison with the conventional method, and increases accuracy by about 5_??_10 times.