Emotion recognition and evaluation from Mandarin speech signals

The exploration of how human beings react to the world and interact with it and each other remains one of the greatest scientific challenges. The ability to recognize affective states of a person we face is the core of emotional intelligence. In the past, several classifiers were adopted independently and tested on several emotional speech corpora with different language, size, number of emotional states and recording method. This makes it difficult to compare and evaluate the performance of those classifiers. In this paper, we implemented a weighted discrete k-nearest neighbor (weighted D-KNN) classification algorithm and compared it with KNN, M-KNN and SVM classification methods by applying them to a Mandarin speech corpus. This speech corpus contains of five basic emotions: anger, happiness, boredom, sadness and neutral. The results of experiments and McNemar's test revealed that the implemented weighted D-KNN method performed best among these classifiers and achieved an accuracy of 81.4%. Besides, we implemented an emotion radar chart which is based on weighted D-KNN and can present the intensity of each emotion component in the speech in our emotion evaluation system. Such system can be further used in speech training, especially for hearing-impaired to learn how to express emotions in speech more naturally.