Summary

In the domain of computer-human interaction, my research addresses the issue of expressivity in voice substitution by synthesis. In situations where a speaker expresses him/herself through a synthetic voice (either singing or speaking), the main challenge is the development of synthesisers that must allow the transmission of at least as much expressive information as natural voice. Current limitations of such systems are twofold: on the one hand, the best performing voice synthesisers mainly focus on intelligibility and do not allow fine control of prosodic parameters that are essential for expressivity. On the other hand, in a context where voice synthesis is used in interaction, the speaker expressivity varies with time in a complex manner, and needs to be transmitted in real-time to the synthesiser. To address both concerns, my work is based on two research axes: the improvement of current voice synthesisers and the introduction of a gestural control, for an exhaustive manipulation of prosodic parameters that allow a fine and subtle transmission of speaker expressivity.