Assessing Teachable Machine by deploying American Sign Language system

Sign language recognition and detection using the current technologies has become one of the focused areas of research which aims to help with communication capability and accessibility for the people of the deaf and hearing issues related community. Over the year with the improvement in the object detection space, A technology by Google ‘Teachable Machine’ which is a web-based tool that allows users to custom train machine learning models with the user’s available data, thereby acting as an automatic ML modeler for producing data models with ease. This paper proposes a method to evaluate the Teachable Machine’s effectiveness by utilizing the American Sign Language (ASL) to make the model recognize the signs once trained. The project involves data collection, image preprocessing, model training using the Teachable Machine’s and testing. The results demonstrated an overall accuracy in predicting ASL letters, though the model is struggling to predict certain letters with similar symbols. The paper also proposes the real-life applicability of such systems in various domains, strengths, and limitations of teachable machines with future scope for improvements in accuracy and real-time processing.