Think to Speak - A Piezoelectric-EEG system for Augmentative and Alternative Communication (AAC) using Recurrent Neural Networks

Abstract

The collection of individuals with severe speech and physical impairments (SSPI), is the target audience for the Think to Speak Augmentative and Alternative Communication (AAC) system. The slow communication rate of AACs accessible to the target audience renders them undesirable, exhausting to operate, and a barrier to social and economic inclusion. This research synergizes the use of Electroencephalography (EEG) and high sensitivity piezoelectric sensor readings with a Long Short-Term Memory Recurrent Neural Network (LSTM RNN) to create a physically accessible AAC with performance comparable to 7.8 characters per minute communication rate. Since self-expression is inextricably linked with physical, mental, and emotional health, this research is of great significance to the estimated one percent of the global population with complex communication needs.