Speaker Independence in Lip Synchronization of Vowels and Distinguishing between /m/ and /

摘要

This paper presents a method for achieving speaker independence in the lip synchronization of vowels by performing an affine transformation on the moments of the spectrum of the speech. A speech sample is separated into samples by glottal pulse period. A track in moment space is made by taking the first and second principle moments of the spectrum of each sample. For a given speaker, a transformation can be created to map the speaker's moment space to a "standard" moment space from which the mouth shape predictor surfaces are created. Also, we describe an approach to distinguishing between the /m/ and / sounds by adjusting based on the velocity of the track moving into or out of the /m/ and / region in moment space. Results are presented for distinguishing "ahm" and "ahn".