부산대학교 도서관

This paper discusses an amusement content in which a robot dances in response to an electric guitar sound played by the guitarist. In our system, the motion of the robot changes according to the guitar performance, and simultaneously, the performer changes his ad-lib by receiving the stimulus from the robot motion and its acoustic percussion sound. In other words, our system realizes a so-to-speak jam session that can give an interactive music stimulus between the performer and robot. The main technical subject of this research is a method of determining an appropriate function from the guitar sound to the robot motion. Our system can extract various guitar sound features from the electric guitar sound. Based on the features, appropriate motions are selected from a motion database prepared in advance. The selection is based on the evaluation result of the audience who have seen the performance. Specifically a neural network that can output the correct answer is constructed by giving a set of highly evaluated sound and motion as training data. We introduce the results of experiments with a 6 degree-of-freedom animal robot holding a maracas by the hand.