Towards Data-Driven Cognitive Rehabilitation for Speech Disorder in Hybrid Sensor Architecture

Recently, research in human-focused robot technologies has aimed to bring closer social-emotional intelligence, which interconnects with human beings' lifestyles doubtlessly. Among several major areas that can benefit from it, the healthcare sector consisting of many possible areas of therapeutic interposition is the most vital one. Recent developments in machine learning have broadened the scope of this work. The present research aims to achieve cognitive rehabilitation for speech disorders using a machine learning approach. Data collected from Electroencephalogram and Kinect sensor has been passed through Chebyshev filter for noise removal, and Autoencoder has been applied to extract features. Finally, Convolutional Neural Network-based transfer learning has shown an accuracy of 96% in classifying speech disorders. The performance of our proposed framework has been quite promising for applying it in real-time scenarios and achieving a better quality of life.