What happens when a virtual character moves like you and speaks when you speak, but the voice you hear is no longer your own?
In this project, the student will create an interactive virtual mirror, where a participant sees a virtual human character responding to their movements and speech in real time. Emerging AI voice-conversion techniques will be used to transform the participant’s voice while preserving what they are saying, allowing the virtual character to speak with a different vocal identity.
The student will investigate current approaches to real-time voice conversion and explore which characteristics can be reliably manipulated, such as speaker identity, apparent age, pitch, vocal style or expressiveness. The project will also address practical challenges such as latency, audio quality and synchronisation between speech, facial animation and the virtual character.
A perceptual experiment will investigate how changing the voice affects identification with the virtual character and the sense that the voice belongs to oneself. Conditions could vary how similar the transformed voice is to the participant’s own voice, or whether the voice is congruent with the appearance of the virtual character.
The project combines generative AI, speech processing, virtual humans, real-time animation and human perception, and would suit a student taking a Machine Learning module with strong programming skills.