Abstract:
For current speech technology, the emotion expression between machines and humans is made possible by dialog modelling (semantics) and acoustic modelling (prosodic features) of speech. Prosodic features alone are considered sufficient to express and perceive primary emotions. With current focus on social robots, there is also the need to synthesize and recognize nuanced secondary emotions. As the secondary emotions are subtle, this study aims to quantitatively assess (via syllable-level prominence features) whether both semantics and prosodic features contribute to their production and perception. Observations show that the effect of semantics on the prosodic features are significant for the production of the secondary emotions. But unlike primary emotions, there is a need for lexical and grammatical information to support the prosodic component enabling people to perceive the secondary emotions. Additionally effects of English language familiarity have been analysed based on the results of a large scale human perception experiment. Keywords: secondary and primary