This work focuses on the effects of context on tone and intonation.
This research develops a broader-context, articulatorily-motivated model of tone, utilizing a common framework across a range of language and tone typologies including Bantu languages, Chinese dialects, and English. Through unsupervised and weakly supervised machine learning, this work aims to automatically identify tone and pitch accent in natural speech.
My work further aims to understand and exploit the synergistic interaction of discourse structure and spoken intonation. To this end, I have developed computational techniques that can automatically detect topic changes in monologue and dialogue as well as corrections in human-computer spoken interactions.
The improved techniques for modeling and recognition of tone and intonation developed in this work will allow computational spoken language understanding systems to more fully exploit the information carried by pitch.