SignBridge: A Real-Time Sign Language to Speech and Text Translation System for Accessible Communication

Despite significant advances in computer vision and deep learning, real-time sign language translation remains constrained by limited vocabulary coverage, high computational cost, and weak integration between recognition and language generation. This study introduces SignBridge, a real-time sign language to text and speech translation system that combines lightweight visual processing with sequence-based deep learning to enable accessible communication between signers and non signers. The proposed system captures live video input and extracts hand landmarks using MediaPipe, transforming raw visual data into structured temporal sequences. These sequences are modeled using a Long Short-Term Memory (LSTM) network to learn motion dynamics and classify gestures within a predefined vocabulary. To address instability in real-time inference, the system incorporates confidence filtering and temporal smoothing, ensuring consistent predictions under varying input conditions. Recognized gestures are subsequently mapped to textual output and synthesized into speech, forming an end-to-end multimodal communication pipeline. The system was developed and evaluated using a subset of the WLASL dataset, which provides diverse, realworld signing conditions across multiple users. Experimental observations demonstrate that the proposed approach achieves stable recognition performance and low-latency response suitable for real-time interaction, while maintaining computational efficiency through landmark-based feature representation. However, the system currently operates at a word-level classification stage and does not fully capture the grammatical complexity of continuous sign language. This work delivers a real-world, scalable way to interpret sign language in real time, focusing on the balance between accuracy, speed, and how easily you can put it into practice. By combining gesture recognition with instant text and speech output, it pushes assistive communication technology forward. It also points out important paths for future research, like nonstop translation, blending different types of signals, and handling bigger vocabularies.