An Artificial Intelligence-based Sign Language Recognition using Attention-Enhanced Long Short-Term Memory Networks

Sign language recognition is essential for enabling communication between the deaf-mute community and the broader population, helping to bridge significant communication barriers. Current techniques tend to use expensive sensors or custom hardware, but this research provides a low cost, AI approach to sign language recognition using video input. An AI model has been proposed in this work that utilizes a self-curated dataset of 14 essential gestures, designed to capture a wide range of commonly used expressions, and employs an LSTM network enhanced with an attention mechanism to capture temporal dependencies. This method allows the model to concentrate on the key frames in the gesture sequences, and can recognize much better, without having to have a lot of processing power. The proposed model achieved an accuracy of 92.1% across 14 gesture classes, demonstrating the feasibility of an efficient and accessible sign language recognition system. This work provides a foundational step toward accessible communication tools for the deaf-mute community, encouraging broader adoption through mobile and low-cost devices.