Voice quality manipulation for humanoid robots consistent with their head movements

This paper presents voice-quality control of humanoid robots based on a new model of spectral envelope modification corresponding to the vertical head motions, and left-right sound-pressure modulation corresponding to the horizontal head motions. We assume that a pitch-axis rotation, or a vertical head motion, and a yaw-axis rotation, or a horizontal head motion, affect the voice quality independently. Spectral envelope modification model is constructed based on the analysis of human vocalizations. Left-right sound-pressure modulation model is established through the measurements of impulse responses using a pair of microphones. The experiments are carried out using two humanoid robots HRP-2 and Robovie-R2. Experimental results show that our method presents the change in the voice quality derived from pitch-axis head movement in a robot-to-robot dialogue situation when the interval between the robots are 50 cm. It is also confirmed that an observable modulation in the voice quality declines as the distance between the robots becomes large. The voice-cast directionality caused by yaw-axis rotation is observable using our model even when the robots stand as far as 150 cm away.