Interacting with Global Content

Research Goals

New methods are being developed to process both speech-only and audio-visual data, and train statistical engines to infer attentional state. The ability to track user engagement and interest in conversational interaction is key to reproducing natural interactions in the future, be that with a robot, personal assistant or avatar. The research explores what makes an avatar, or computer generated speaker, engaging to a user. The research uniquely combines ADAPT expertise on expressive synthesis, the role of paralinguistic cues in speech, and avatar animation.

Also addressed are issues of multimodal content relevant to interaction from two perspectives: The first addresses challenges of locating and isolating objects of interest in a visual stream and exploiting visual cues in speech to augment speech recognition capabilities. Learning techniques from unstructured multimodal data streams are also examined. The second is addressed by establishing methods for the exploitation of dialogue in user interaction in information retrieval, and to the exploitation of context to enable proactive information retrieval.