Most of speech signals exhibit a discrete power spectrum: i. e. a limited number of values only are of importance, which generally correspond to harmonic partials of voiced sounds. It is well known that source-filter modeling of such signals leads to difficulties with usual methods such AR modeling. This article deals with a new method for modeling this class of signals which extend and unify preceeding works.
Bibliographic reference. Galas, Thieny / Rodet, Xavier (1991): "Generalized functional approximation for source-filter system modeling", In EUROSPEECH-1991, 1085-1088.