Speech Emotion Recognition and Sentiment Analysis based Therapist Bot

Access to mental health care and treatment is a global concern. The demand for these services outnumbers the supply. Online treatment delivered by a chatbot might not only provide access to cost-effective aid, but it could also be convenient for individuals, who are hesitant to participate in therapy. The main aim is to leverage technology to enable more people to seek help. The proposed method has built a responsive therapist bot that generates appropriate responses according to a person's emotion. The detection phase uses an ensemble of deep learning models for Speech Emotion Recognition (SER) and text based sentiment analysis, which classifies one's emotions into four categories – happy, sad, angry, and anxious. The proposed approach can be proved to be fool-proof and it is better than existing methods due to the ensemble of two efficient models CNN (83%) and BiLSTM (92%). In addition to detection and responses, this application recommends suitable tasks to assist the target audience.

I. Introduction

As per statistics from WHO, nearly 7.5 % of Indians suffer from different mental disorders. It is predicted that by the end of this year, close to 20 percent of India will suffer from mental illnesses. While anxiety and stress majorly affect people in the 15–29 age group, most of them do not seek professional help. This could be due to financial issues or mental health related stigma. The aim of this paper is to automate this process by creating an effective and fool-proof method to detect emotions. Therapist Chatbots have gained a lot of popularity over the recent years. However, existing systems in the market are text-based and use techniques like Cognitive Behavioral Therapy and meditation. While these have proven to be helpful, human emotions change almost every second and it is very challenging to articulate strong emotions into words. In fact, some emotions often go unnoticed in text-based systems.