A machine learning-based system to translate algerian sign language to speech

Abstract

This thesis presents the development of a real-time Algerian Darija sign language translation tool, which was thought to facilitate the communication of hard-of-hearing and deaf communities.The said tool recognizes Arabic letter hand movements and converts them into verbal Darija through pre-recorded sound files. Implemented in Python, the system combines MediaPipe’s hand- tracking library with a convolutional neural network (CNN) that has been trained on a bespoke gesture dataset. Designed to run offline on a Raspberry Pi Zero 2W , it uses pre-recorded Darija audio files to vocalize recognized gestures. Testing showed high accuracy under the specified environment and the assigned hardware Its ease of use and modularity allow for future expansion, with possibilities for mobile integration . This study adds to the body of assistive technology by offering a low-cost and contextually relevant solution that facilitates communication between sign language users and nonsigners in daily communication.