The relationship between voice and face: A cross-modal approach to talker identity

Humans have developed a remarkable ability to recognise individuals through voice and face. There also seems to be a mysterious link between voice and face - we seem to have a mental calculation to envisage what a person might sound like when we hear a voice, or speculate what a person might sound like when we see a face. We would be surprised if the real face or voice turned out to be too different from what we imagined. A recent success in deep learning technology allows for high-precision predictions of faces from voices based on huge data sets collected from the web. Human and machine performance indicates a strong interconnection between voice and face. However, we still have very little knowledge of the fundamentals of the link between the two domains, hence the motivation for this project.