Adaptive Hierarchical Emotion Recognition from Speech Signal for Human-Robot Communication

Abstract

Emotional speech recognition is an interesting application that is able to recognize different emotional states from speech signal. In Human-Robot Interaction (HRI), emotion recognition is being applied on intelligent robots so that they can understand emotional states of user and interact in a more human-like manner. However, it is not easy to apply emotion recognition algorithms in real applications due to the dependence on many factors. In this paper, we introduce hierarchical approaches that generate the binary classification tree automatically and exploit multiple classifiers to recognize different emotions. And then we propose a framework that recognizes emotions from speech signal with a higher accuracy and efficiency in comparison with other algorithms such as Hidden Markov Model (HMM) or Support Vector Machine (SVM). The method automatically creates a binary classification tree and optimizes the classifier at each node of this tree so that the recognition result will be achieved with a higher accuracy and performance. The recognition phase is simple to implement on different mobile platforms with less computational efforts than other approaches.