Abstract:

We developed an anthropomorphic talking robot WT-4 (Waseda Talker No. 4) to produce human speech. WT-4 consists of 1-DOF lungs, 4-DOF vocal cords and articulators (the 7-DOF tongue, 5-DOF lips, 1-DOF teeth, nasal cavity and 1-DOF soft palate); the total DOF is 19. The lips and the tongue are made of elastic material to allow large deformation and to prevent from the air and sound leaks, and are controlled by a looped wire mechanism to form their various configurations. The talking robot enables to produce vowel and consonant sounds by mimicking the vocal cords vibration and the fricative and plosive source generation by the air flow as well as dynamically controlled vocal tract acoustic resonance in human speech production. Articulatory control of the talking robot is designed to track the acoustic goals (pitch, sound power, two formant frequencies, and voice-unvoiced timing) of the speech. The robot parameters are determined by minimizing the weighted mean squared error of these acoustic parameters between the human and robot speech sounds. It is shown that this mimicking speech control is effective in producing fluent continuous speech by the talking robot.