Gesture Talk -Bridging Silence with Words

This is the authors' abstract. We don't add key points for this paper.

Gesture Talk is a real-time sign language-to-text conversion to create a communication environment for the deaf and able-bodied, with a touch of inclusivity across many other domains. The whole idea is to further develop a cheap and efficient solution to translate American Sign Language (ASL) gestures into text without the necessity for an interpreter. The process uses convolutional neural networks (CNNs) with a dual-layer classification algorithm, applying Gaussian blur and adaptive thresholding on the pre-processed webcam video frames and incorporating an auto-correct feature to enhance word prediction. The system was trained and tested on a labeled ASL gestures dataset prepared out of pre-processed 128×128 grayscale images obtained from RGB video inputs. Gesture Talk achieves recognition of 98.0% accuracy for ASL gestures, which surpasses many existing systems, and provides a user-friendly interface supporting all platforms, enabling deployment on desktops, mobiles, and web applications for much greater accessibility for deaf individuals.