Authors: Samruddhi Pramod Bangar, Prasad Sakharam Khose,, Nikita Sandeep Ubhe, Sonali Dongare
Abstract: Communication barriers persist for hearing and speech-impaired individuals due to the lack of real-time, offline, bidirectional Indian Sign Language (ISL) translation tools. This paper presents SignBridge, an offline bidirectional system supporting Sign-to-Text/Speech and Text-to-Sign translation. MediaPipe is used for extracting hand and pose landmarks, while a TensorFlow-based hybrid CNN– Transformer model performs dynamic gesture recognition. A full-stack implementation using React.js and Flask ensures real-time interaction, and an avatar-based rendering module generates visual sign outputs. The system is designed for low- resource environments with improved privacy and reduced dependency on internet connectivity.
