Nonlinear interpolation of speech-spectra using soliton equations

Abstract

This research is devoted to the study of how to get an intelligible speech synthesized from a pair of speech-spectral sequences YA and YB uttered the same sentence by a speaker A and B. First we relate a certain smoothness of vector fields of a surface with the boundary of YA and YB to the intelligibility of synthesized speech-spectral sequences. Second we show experimentally that a flat connection of the surface derived from soliton equations play an important role in maintaining speech intelligibility.