The right kind of unnatural: designing a robot voice

Any system using voice to communicate becomes personified by that voice. For robots, where the form and non-vocal behaviour also strongly personify the system, we can see a clash between the two technologies. The many challenges in building responsive and interactive robots mean that language systems are often designed in a vacuum and when they are finally brought together can ruin the look, feel and sound of the completed system. This problem is intensified by natural language processing technology which can further add inappropriate behaviours following mythical business use cases, rather than exploring how users really would like to relate and use embodied artificial systems. In this positional paper, we present two studies in robot voice design and a non-vocal use case of the Honda Research Institute robot Haru. Finally, we ask the question what sort of voice should Haru have?