AI-Powered Sign Language Translation System: A Deep Learning Approach to Enhancing Inclusive Communication and Accessibility in Low-Resource Contexts

Abstract

This paper addresses the persistent communication gap between Deaf and hearing communities by presenting an AI-powered, real-time sign language translation system. The proposed system employs a hybrid deep learning architecture that combines Convolutional Neural Networks (CNNs) with Transformer models, enhanced by advanced keypoint detection technologies such as MediaPipe. This configuration enables the translation of continuous sign language video into natural language text and speech, capturing the complex spatiotemporal dynamics of signing gestures. The system is designed with scalability and inclusivity, featuring a modular full-stack architecture deployable across diverse platforms. A particular focus is given to low-resource sign languages, which are frequently marginalized in current technological solutions. Quantitative evaluations demonstrate notable improvements in translation accuracy and user experience compared to traditional recognition systems. In addition to technical innovation, the study emphasizes ethical design principles ensuring cultural sensitivity, user privacy, and community-centered development. The findings contribute a scalable and practical solution that enhances accessible communication while promoting global health equity and social inclusion. This work also underscores the value of interdisciplinary collaboration and policy alignment in developing equitable assistive technologies.

Keywords: AI, Artificial Intelligence, Blockchain