Enhancing Sign Language Recognition for Hearing-Impaired Individuals Using Deep Learning
R. Arularasan, V. Balaji, Srisudha Garugu, Venkateswara Rao Jallepalli, S Nithyanandh, Gokulakannan Singaram · 2024
Communicating with people who have developmental disabilities and hearing difficulties requires the use of sign language. It presents the SSODL-ASLR model, which is customized for people with speech and hearing impairments, by using advances in CNNs and HCI. The model works in two stages: first, it uses the Mask RCNN model for sign language detection, and then it uses an SSO approach and the SM-SVM model for sign language classification. Prolonged simulations demonstrated significant improvements with the suggested SSODL-ASLR model: 15% better accuracy, reaching 95%; 92% precision; 94% recall; and 93.5% F1 score. The model also achieved a reduction in memory footprint to 250 MB, a processing speed of 120 FPS, a training time of 12 hours, and an inference time of 20 ms. These results demonstrate the revolutionary effect of the SSODL-ASLR paradigm in promoting automated sign language recognition and enhancing the deaf and mute community's communication accessibility.