SER Analysis -An Approach to enhance Human Health

This work focuses on Speech Emotion Recognition (SER) for extracting emotional states from speech signals, encompassing universal emotions like Happiness, Anger, Neutral, Calm, Sadness, Disgust, Fearful, and Surprised. The methodology involves extracting features such as Chromogram, Mel- frequency cepstral coefficients (MFCC), Mel scaled spectrogram, Tonal Centroid and Spectral contrast, followed by classification using Artificial Neural Networks (ANN). A health-oriented aspect, is integrated in this model by providing personalized guided meditation sessions based on the user's emotional state to promote mental wellness. The main aim of this work is to broaden its reach and impact through ongoing refinement of the model and exploration of new implementation strategies, ultimately contributing to a healthier and happier society. The results of this work demonstrate promising accuracy, with the trained deep neural network (DNN) model achieving approximately 81.25% accuracy in recognizing speech emotions in comparison with the existing algorithms. DNN shows a 11%, 8%, 14% increase in accuracy with RNN & LSTM, SVM, GMM respectively.