Rhythm correlation of speech synthesis system

Abstract

There has been a rapid progress in speech synthesis; however, it is still hard to make a good objective evaluation of speech intonation while training the speech synthesis system. Unlike the traditional method, standard deviation of intonation, which normally makes the speech synthesis system sound smooth and flat, but with less expressiveness, the paper integrates the rhythm correlation in an evaluation based on tangential intonation. Furthermore, the paper makes a comparison among three typical evaluation methods: listening test; standard deviation of intonation; standard deviation of intonation and tangential intonation. It proves that the introduced method could generate better synthesis results than others with an even smaller training corpus.