Machine Learning Driven Musical Improvisation for Mechanomorphic Human-Robot Interaction

摘要

As industrial robots and social robots become prevalent in commercial and home settings it is crucial to improve forms of communication with human collaborators and companions. In this work, I describe the use of musical improvisation to generate emotional musical prosody for improved human-robot interaction. This aims to develop a canny approach, where robots perform in a mechanomorphic manner improving collaboration opportunities with humans. I have currently collected a new 12-hour dataset and developed a Conditional Variational Autoencoder to generate new phrases. Generations have then been used to compare the impact of prosody on anthropomorphism, animacy, likeability, perceived intelligence, and trust. Future work will incorporate prosody into groups of robots and humans, using personality to drive emotional decisions and emotion contagion.