Incremental learning for spoken affect classification and its application in call-centres

Abstract

This paper introduces a system for real-time incremental learning in a call-centre environment. The classifier used is a Support Vector Machine (SVM) and it is applied to telephone-based spoken affect classification. A database of 391 natural speech samples depicting angry and neutral speech is collected from 11 speakers. Using this data and features shown to correlate speech with emotional states, a SVM-based classification model is trained. Forward selection is employed on the feature space in an attempt to prune redundant or harmful dimensions. The resulting model offers a mean classification rate of 88.45% for the two-class problem. Results are compared with those from an Artificial Neural Network (ANN) designed under the same circumstances.