Synthesising and Evaluating Cross-Modal Emotional Ambiguity in Virtual Agents

摘要

Emotional ambiguity, when more than one emotion appears present at a given time, or several emotions are superimposed, is common in human interaction and effects such as irony can be intentionally created through a mismatch of such emotional signals. High quality emotional speech synthesis offers a means for testing the effect of combining differences in vocal emotion, facial expression and text content in a virtual agent. In this paper we combine high quality emotional speech synthesis with a video rendered non-naturalistic virtual agent. Vocal emotion and text content combined to increase or decrease the emotional valence (positivity) of an utterance, while emotional facial expressions did not affect valence, but interacted with vocal emotion altering emotional activation in the lax and stressed vocal condition.