AI Based Web App and Framework for Detecting Emotions from Human Speech

It has been a challenging task to identify meaningful clues in speech/voice data to recognize emotion. Speech Emotion Recognition (SER) system is a collection of methodologies that process the speech data, extract useful features, use efficient classifier/classifiers to finally detect the embedded emotion from the six basic emotions, i.e., happiness, surprise, anger, sorrow, disgust and fear available. Increasing applications of SER are found in health care system, security, entertainment, education, BPOs, marketing, defense, customer care and Human Computer Interface (HCI) systems. This paper focuses on building a web-based interface for recognizing emotion from an audio file uploaded by user. The web app for SER is built on the flask framework following the supervised learning approach using Python. The system has been tested on publicly available TESS dataset with both Machine Learning (ML) and Deep Learning (DL) based models. Performance comparison of all the trained models is done with state of the art. Results are quite promising pointing towards developing a robust SER system.