Speech emotion recognition

Abstract

Speech is one of the most natural ways for humans to express themselves. We rely on it so much that we notice its relevance while using other modes of communication, such as emails and text messages, where we frequently utilise emoticons to describe our feelings. Because emotions are so important in communication, detecting and analysing them is critical in today’s digital age of remote communication. Because emotions are subjective, detecting them is a dif ficult task. There is no universal agreement on how to quantify or classify them. In this Project, we used the most widely used audio features like MFCC, MEL spectrogram and Chroma to classify each and every emotion. We test our algorithm’s performance with different models to find the best fit for our use case. The dataset used in the Project is RAVDESS which is by far the best dataset available for the underlying problem of emotion detection. It contains in total of 1440 samples of audio detection. Atlast, we have shown our results with accuracies of different models, their confusion and classification matrix.