Individual control of singing voice based on cepstrum manipulation

Abstract

Desktop music (DTM) software is used to synthesize various instrumental sounds, whereas it was difficult to synthesize the natural singing voice because the singing voice is more complicated than other instrumental sounds. In the past, it had been object that we output more natural singing voice by analyzing the singing voice and extracting the spectrum envelope with high accuracy. Recently, since the algorithm is improved, the singing voice synthesis software is used. In the conventional method, it had been focused changing the original voice personality by controlling some parameters of spectrum envelope. However, the method can synthesize the singing voice by only particular voice personality. The quality of the synthesized singing voice also depends on accurate control of voice personality. In this study, we attempt to control user’s impression directly and investigate the control method that is focused on the singer's individuality in the singing voice. In this paper, we propose a voice personality control method based on mapping the timbre of target singer in the cepstrum domain and demonstrate the effectiveness of the proposed method. As a result of subjective experiments, we confirmed that the proposed method can control the voice personality of singers with high quality.