Speech based Emotion Recognition using Machine Learning

Abstract

Emotions help a lot in recognizing the feelings of a human being. As per the study, there are multiple ways such as Linguistic, Video, Physiological signals, Audio cues, etc. that help in analyzing emotions. In this paper, analysis of speech has been done as it is the most natural way of showing emotion. For the experiment, RA VDESS (Ryerson Audio-Visual Database of Emotional Speech and Song) database is used that contains audio files of various people that represent discrete emotions. Based on speech data, a comparison of various machine learning classification algorithms has been done that assist in the categorization of multiple emotions. The compared algorithms use some classifiers that are Recurrent neural network (RNN), Support vector machine (SVM), K- Nearest Neighbors (k-NN), Adaboost, Gradient Boosting Classifier, Multi-Layer Perceptron (MLP), and Random Forest. The classifiers were analyzed extracted features such as Mel Frequency Cepstral Coefficients (MFCC), Chroma, Mel: Mel Spectrogram Frequency, Spectral Contrast, and Tonnetz available in librosa library of python language. After analysis, experimental results reveal that among all other classifiers most accurate approach for emotion recognition with 89.5% was achieved by MLP classifier.