A study on the consistency of human perception and machine recognition of an emotional corpus

Emotion plays a critical role in human interaction. Listeners can perceive the emotional state of speakers from their facial expression, gestures and/or speech. In this paper, we investigate the relationship of the intended emotion expressed by speakers and the emotion perceived by listeners of a newly recorded corpus. We investigate the consistency of the emotion expressed by speakers and the emotion perceived by listeners. Furthermore, we also compare the results with the outcomes of a speech emotion recognition system. The results indicate that females have a better ability in expressing emotion than males. However, females have greater variance in perceiving emotion. Another finding is that the recognition rates will increase as the duration of an utterance getting longer in length.