On the use of pitch-based features for fear emotion detection from speech

In this paper, we present a study that aims to evaluate the effect of pitch-related features on fear emotion detection from speech signal. In this context, several features have been tested. Only relevant ones are selected thanks to ANOVA tests. Next, they were decorrelated using principal component analysis. To select fear, emotion classification based on machine learning methods is used to extract fear from other emotions. Many classification tools are used and compared. We considered two types of emotion classification which highlights the fear emotion state, a simple classification as well as an hierarchical one. Results show that selected pitch-based features have a relatively great power in fear recognition. In fact, the highest accuracy rate reaches 78.7% using k-nearest neighbors algorithm.