Title: Human Speech – How Do I know it’s Real?
Description: How can you tell when speech is real, or when it is fake? This is the focus of this PhD project and it goes to the very core of the nature of human speech. Directly relating what is observable at a signal level in speech to how natural that signal is, as perceived by a human, is an unsolved problem in speech technology. This PhD addresses this gap in knowledge. The research will leverage a wealth of data from recent Blizzard speech synthesis challenges, where the naturalness of multiple world-class speech synthesis systems has been rated and made publicly available for researchers. Simultaneously, the research will also leverage shared datasets on spoofing from the automatic speaker verification community, such as those available through http://www.asvspoof.org/. The research is truly novel in that it goes beyond treating speech purely as a signal, and will bring the work to the level of investigating naturalness in continuous speech, over many seconds and sentences of generated speech.