Speech Emotion Recognition Using Multi-Layer Perceptron Classifier

This paper proposes a speech emotion recognition approach using the Multi-Layer Perceptron Classifier (MLP Classifier). The Mel-Frequency Cepstral Coefficients feature and openSMILE feature are respectively extracted. With the extracted features, MLP Classifier is used to classify the speech emotion. The Berlin database which contains seven emotions: happiness, anger, anxiety, fear, boredom and disgust, is used to evaluate the performance of the proposed approach. Data augmentation are furtherly employed and experimental results show that the proposed approach achieves satisfied performances. Comparisons are conducted when with data augmentation and without data augmentation, and the results indicate better performance with data augmentation.