An Automatic System for Recognition and Assessment of Anger Using Adaptive Boost

This paper proposes a method to identify and assess different levels of anger from the speech utterances. Unlike the existing methods which only detect the emotion from speech, the proposed method not only detects but also labels the level of an emotion. A 75 dimensional feature vector has been extracted from each audio clip and is used for training and testing. For classification and assessment the adaptive boosting algorithm is used. Experiments were performed on a dataset of seven emotions. 231 audio clips were used for training and 100 were used for testing. The accuracy of the proposed system to detect 'Angry' emotion is 78.3%. All the angry clips are classified into low, medium and high level of anger.