The ability to capture and reproduce the voice timber and the speech prosody of a given person has
made impressive progress, e.g., with recent Ircam projects: based on several hours of recording of an
actor reading texts, entirely new synthetic speech of incredible resemblance can be uttered in the
same language; also, using much less training data, voice timber can also be transferred to some
extent to the natural speech of another speaker. In this work, we would like to investigate ways to
push further such systems to create polyglot voice doubles, with various applications. The intern will
have to get acquainted with existing voice analysis and reproduction (either by synthesis or by
transfer) techniques and will try and alleviate the current mono-lingual limitations.
Keywords:
Voice analysis, Prosody, Speech Synthesis.