Aiding therapy using speech emotion recognition

• Main Author: Koh, En Rong
• Other Authors: Qian Kemao
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: https://hdl.handle.net/10356/153290
• Institution: Nanyang Technological University

In the past 20 years, mental health has come to light within society. The stigma surrounding mental illness is declining thanks to the increasing awareness and encouragement through social media and digital platforms. The growth in psychologists and therapists can also be seen in recent years. Not only did the mental health industry has an increase in patients and counsellors, the advancement of technology integrating with this field is visible in the present day of mental health care. It brought a significant impact on aiding individuals deprived during this period of time. Research on artificial intelligence also improved the quality of therapy, bringing it closer to people who are struggling and taking over virtually. Nonetheless, the applications need to be carefully designed and balanced against their limitations, depending on different mental illnesses. While different kinds of AI have been assisting in the mental health field, such as therapy chatbots and virtual therapists, a lack of recognizing human emotions can be commonly seen in AI systems, especially through speech. Speech Emotion Recognition became a research topic in a wide range of applications and became a challenge in speech processing. In this project, an AI Speech Emotion Recognition system is experimented with using Deep Learning techniques to alternative traditional methods like Support Vector Machine or Hidden Markov Model. We will explore the use of a Convolutional Neural network, a type of Deep Learning method, to train and predict human emotions. We will also examine the different types of time-frequency features in audio signal processing and how they help in classifying human emotion. A SER system with a visual modality will also be developed to test on real-time prediction.