Design of a System of Response Options Using a Neural Network for People with Communication Difficulties

Abstract

The paper proposes a communication system aimed at improving the interaction of people who have lost the ability to speak due to various circumstances, such as accidents, facial paralysis, or vegetative states. The system consists of speech recognition and a neural network with Bidirectional Encoder Representations from Transformers (BERT) structure, which was trained to provide response options in order to establish fluid and meaningful interactions and maintain a conversation between the patient and the environment. Its design seeks to improve the quality of life of those with substantial communication challenges; for this purpose, we propose an interface consisting of two modes, one for people who have some mobility in their hands and another for people in a completely vegetative state using Electrooculogram (EOG) systems to configure the signals with our proposed system. The effectiveness of the system was evaluated using the subjective Scale Utility System (SUS) method, yielding promising positives that underline its effectiveness and perceived usefulness for real environments.