Valence-arousal approach for speech emotion recognition system

The current state-of-the-art speech emotion recognition approaches focus on discrete emotion classification to suit the users' need. However, in more practical perspective, emotion is deemed complex to be individually segregated and it is a continuous process that will change dynamically over time. Subsequently, no researcher can really claim of being able to find the threshold discriminating one emotion to another. In this paper, an alternative approach of Valence-Arousal is introduced based on the psychologists' understanding that emotion can be represented in two emotion primitives of Valence and Arousal. Empirical results show potential of the proposed approach, which achieved reasonable accuracy with better speech emotion understanding and analysis.