Learning Internet from tone of voice

Abstract

The ability to discern information from the tone of voice that a person uses is an important part of social interactions. Synthetic characters that can interact naturally with humans could take advantage of this information if they could discern it. I propose that a synthetic character with a vocalization affect classifier and the ability to learn associations can use the tone of voice of the person interacting with her to predict what the person is going to do. In this approach the classifier learns to distinguish tones in real time allowing the character to adapt to new tones. I describe the implementation of the system, called Minimus T.O. Mouse, and its extensions from previous affect classifying systems and previous synthetic characters.