A Software based ASL Translator and ASL to Voice Emulation System

Abstract

Sign languages use the visual modality to convey meaning. People who are hearing and speech impaired are left behind video consultations as there is no easier way for non vocal people to communicate with vocal people and express themselves to the hearing world. Since they can't sign themselves they are forced to use a basic text chat to hold their consults. To overcome this issue, we have developed a real time software based ASL Translator and voice emulation system where the image captured is converted into predicted text. We use Deep Neural Networks like Convolutional Neural Network for performing feature extraction and the SVM classifier for classification. The system returns top and most probable predictions. We obtained an accuracy rate of 93.75%.