||Synthesis of expressive singing voice by F0, amplitude envelope and spectral feature conversion
Nguyen, Thi-HaoAkagi, Masato
2018 RISP International Workshop on Nonlinear Circuits, Communications and Signal Processing (NCSP2018)
690 , 2018-03-07 , Research Institute of Signal Processing, Japan
This paper investigates correlates of acoustic features to emotional singing voices. By analyzing acoustic features that are strongly related to emotions, this research determines which feature is more significant to the emotional expressions in singing voices. We also propose a method to modify amplitude envelopes based on the entire F0 contour to have a higher naturalness as singing voice. The results show that the spectral feature is the most affecting acoustic feature to the emotion of singing voice. However, in order to obtain high naturalness and singing-ness for the synthesized voices, it is necessary to manipulate all three features that are F0 contour, amplitude envelope and spectral sequences.