Transformative Ai-Driven Tamil Sign Language Recognition and Speech Synthesis Using ViT-CNN

Tamil Sign Language (TSL) is devoid of automated translation software, restricting its access to people with hearing and speech impairment. This paper suggests a hybrid Vision Transformer (ViT) and CNN model for real-time TSL recognition and voice synthesis. The model combines CNN for feature extraction, ViT for learning attention, and BiLSTM for sequence gestures with 97. 5% precision. The recognized gestures are synthesized into Tamil speech using Tacotron 2-based synthesis. A specially designed TSL data set provides robustness with variations in gestures. The system is optimized for low-latency edge device deployment. The results indicate that ViT significantly enhances the accuracy of sign recognition, filling communication gaps for the disabled.