Real-time spoken affect classification and its application in call-centres

We propose a novel real-time affect classification system based on features extracted from the acoustic speech signal. The proposed system analyses the speech signal and provides a real-time classification of the speaker's perceived affective state. A neural network is trained and tested using a database of 391 authentic emotional utterances from 11 speakers. Two emotions, anger and neutral, are considered. The system is designed to be speaker and text-independent and is to be deployed in a call-centre environment to assist in the handling of customer inquiries. We achieve a success rate of 80.1% accuracy in our preliminary results.