Speech Emotion Recognition Using Machine Learning Techniques

Abstract

Mel Frequency Cepstral Coefficient (MFCC) method is a feature extraction technique used for speech signals. In machine learning systems, the Random Subspace Method (RSM) known as attribute bagging or bagged featuring used to classify the complete feature sets. In this paper, an innovative method is proposed which is a combination of RSM and kNN algorithm known as Subspace-kNN (S-kNN) classifier. The classifier selects the specific features extracted from MFCC are angry, sad, fear, disgust, calm, happiness, surprise, and neutral speech emotions in Speech Emotion Recognition (SER) system. Furthermore, in the proposed method the performance metrics of accuracy, Positive Predictive Values (PPV) rate, training time are evaluated on male and female voice signals when compared with previous classifiers like SVM and bagged trees.