Mechanical design of a talking robot for natural vowels and consonant sounds

Vocal movement isn't only a movement of the vocal organs. It is also a movement that produces acoustic signals received by hearing as linguistic information through hydroacoustic phenomena along with the formation of the vocal way. The purpose of this research is to make clear the human vocal mechanism from the view of engineering by simulating the vocal movement with a robot, and to create the dynamic model. Therefore, the authors developed an anthropomorphic talking robot WT-1 (Waseda Talker-No.1) in 1999. It simulates human vocal movement, and has articulators (the 6-DOF tongue, 4-DOF lips, 1-DOF teeth, a nasal cavity and 1-DOF soft palate) and vocal organs (the 1-DOF lungs, 1-DOF vocal cords); Total DOF of the robot is 14. We experimented with it on vowels. As a result, F1 and F2 frequencies of all Japanese vowels were similar to the human averages. WT-1 could utter single vowels. However, its voice wasn't natural. In this paper we describe the improvement of the mechanisms for the realization of nat ural vowels and consonant sounds.