Generation of presentation motion using sequence to sequence model

Abstract

In the presentation, gestures play a very important role as a means of improving the information transmission effect. Moreover, it was shown that the movement expressing the enthusiasm and intention of the presenter in an unconscious manner affects the success of the presentation and the impression on the audience. For these reasons, it is required for presentation robots to perform such movements; however, manual design of these movements is not an easy task. In this research, we propose a method to learn relationship between speech prosodic information and motion using a sequence to sequence model, and directly generate appropriate motions from prosodic information. This paper also proposes a method for generating motions representing meanings of specific words.