Abstract
This is a study on a singer robot using artificial larynx. The robot can generate Japanese 5 vowels [a], [i], [u], [e], and [o], in the present. It is very difficult to generate clear vowels because the frequencies of formants are not easy to be controlled. Formant frequencies can be manipulated by changing the cross-sectional area of the vocal tract. Singer Robot is expected to produce vowels with correct formants in order to sing in distinct pronunciations. We propose a new adjustment system of vowel formant. This system can search an optimal cross-sectional area distribution of vocal tract for the given magnitude of the formant frequency change. And, the system introduced in this paper analyzes formants in real-time and feeds back them to Singer Robot.
