Emotion Detection Through Speech Analysis

In today's day and age of digital assistants there is a whole new avenue of data that is not being tapped, the audio signal that is being spoken to these assistants. This can be used to great effect be a variety of industries that face regular challenges in identifying the emotional makeup of their clients. Institutions like hospitals, emergency service centers would find such a decision support system invaluable in their day to day working. Objective: Create a multi-label classification model that will identify the emotion from speech samples. Methodology: We employ the use of various different classification models and compare and contrast their outputs using robust mathematical evaluation metrics to try and find the most optimal model for the use case. Results: We can see from the table 2 that in this study the Deep Neural Network (DNN) based model performs the best among the various classification models employed with an overall accuracy of almost 78%