A method of creating a new speaker's VoiceFont in a text-to-speech system

Abstract

This paper presents a method of creating a new speaker's voice database (VoiceFont) by which the voice of the donor speaker can be synthesized for mimicking in a text-to-speech system. A VoiceFont creation system, "VoiceFont Builder", is developed to make the creation process easier and more effective than current systems. The voice feature extraction applied in the system is a simple but powerful method that makes the most of the target speech synthesizer. Using a VoiceFont obtained, we conducted experiments on F0 contour generation in view of reproducing that of the donor speaker's voice.