Enhancing Communication for the Deaf and Hard-of-Hearing: A Custom Deep Learning Model-Based Approach for Real-Time Sign Language Recognition and Translation

This research presents a groundbreaking sign language detection and real-time word label translation system, demonstrating a significant improvement in recognition accuracy across multiple datasets. Utilizing a custom-built neural network that integrates advanced deep learning techniques, our system has achieved top accuracy rates, outperforming existing models. For instance, on the WLASL-2000 dataset, our model reached an accuracy of 86.15%, surpassing previous works significantly. The system’s effectiveness was further validated in real-life applications, proving its reliability and utility in facilitating communication for the deaf and hard-of-hearing community. By providing a user-friendly interface and enabling seamless real-time translation, our approach enhances accessibility and inclusivity, offering a transformative tool for better integration of individuals with hearing impairments into society. This advancement not only empowers communication but also promotes equal opportunities in education and employment, contributing profoundly to humanitarian efforts.