Implementing Emotion Detection from Speech for Psychological Assessment of Elderly People: A Comparative Study of Python- based Approaches and Existing Solutions

In the last ten years, the number of people over 65 has increased 30% in Spain. This trend is anticipated to grow and require more healthcare personnel. To prevent this, people should live longer independently instead of in care homes. The ADDIM system will assist them in living independently. The research presented in this paper is part of the mood detection of the user in the ADDIM (Asistencia Domiciliaria Digital Integral para Mayores) system. This is a Digital platform for monitoring older people's health, safety, companionship, and emotional support at home based on robotics, artificial intelligence, and ambient assisted living. To detect user emotions, the right speech corpus, feature extraction methods, preprocessing methods, and machine learning models have to be selected. Based on the detected emotion, the robot will interact with the user to perform predefined actions. The final mood of the user will be estimated using this output in conjunction with visual feedback and the sensors in the user's home with the ADDIM system. Three speech corpora are selected with retraining to achieve personalized detection based on the user's previous recordings. In addition, this will ensure that the detection is improved over time, which has yet to be implemented in other research. Finally, the implementation uses dimensional emotion detection instead of discrete emotion detection. This augments the number of detectable emotions.