A lyrics to singing voice synthesis system with variable timbre

Abstract

In this paper, we present a singing voice synthesis system, which can convert lyrics to singing voice. As the synthetic song's timbre is too monotonous, a new singing voice morphing algorithm based on GMM (Gaussian Mixture Model) was presented accordingly. The MOS test shows that the average MOS score of synthesized song is above 3.3 before timbre conversion. The professional singer's timbre can be added proportionally by changing the scale factor k in the system. The ABX test demonstrates that the accuracy can be up to 100% in the case of k=0 or k=1, and it can be higher than 64.5% in the case of 0