Singing voice synthesis from read speech using STRAIGHT

摘要

This paper proposes a method to transform read speech into singing voice by using STRAIGHT. In order to achieve the method to transform read speech into singing voice, the followings must be required; (1) F0 control based on melody component, (2) duration control of phonemes based on rhythm, and (3) spectrum control based on F0 variation. This paper introduces duration control rules for each type of phonemes and spectrum stretching methods for controlling spectral shapes on frequency axis as a function of the read speech F0 and the melody component, as well as a F0 control model the authors have proposed. Quality of the synthesized singing voice is almost the same as that of sung voice.