Classification of anger emotion using Japanese vowel

Anger emotion is categorized into “hot-anger” and “cold-anger”. In many previous researches, it has been shown that cold-anger is more difficult to be recognized than hot-anger because it has few obvious characteristics. In this paper, we use Japanese vowel “a” to discriminate the three voice classes, “hotanger”, “cold-anger” and “neutral”. In the experiment, four machine learning methods, deep neural network, support vector machine, decision tree and Λ-nearest neighbor, are applied to the logarithmic power spectrums of the vowels collected from five Japanese actors. As a result, it is shown that the three voice classes can be discriminated using only vowel with accuracy comparable to existing approach that utilizes multiple acoustic features.