Speech Emotion Recognition Using Multi-layer Perceptron

In the modern digital era of distant communication, emotion detection and analysis are essential, since emotions are key component of communication. Emotions are arbitrary, making it harder to discern them. There are differing opinions on how to quantify or categorize them. From machine learning perspective, speech emotion recognition is a classification problem where an input sample (audio) needs to be split into a few specified emotions. It is a trivial task to recognize emotions from only speech. In this paper, we made it possible through feature-based approach. This problem is a part of multi-class classification; thus, we used five artisan features extracted from an audio sample. We have trained four models: Support Vector Machine (SVM), Random Forest (RF), Convolutional Neural Network (CNN), and Multi-layer Perceptron (MLP) and reported their accuracy, f-score, and recall for all the machine learning models. Confusion matrices for all the models are also provided. To detect the emotion: pitch, speaking rate, and energy are taken as features. Results have shown that Multi-layer Perceptron, i.e. MLP, has performed the best by providing highest accuracy 93.17%, respectively.