Real-Time Sign Language to Speech Recognition System for BSL

This paper presents a groundbreaking real-time system that transforms British Sign Language (BSL) gestures into audible speech using long short-term memory (LSTM), employing innovative machine learning techniques. The contribution, rooted in a blend of quantitative and qualitative research methods, showcases significant advances in technologies that assist BSL users. By integrating TensorFlow and MediaPipe, we designed a sophisticated multi-layer LSTM neural network that effectively handles real-time translation. Our approach encompassed several stages: gathering data, intensive preprocessing, meticulous training of the model, and thorough testing. The results from these tests confirmed the model's high accuracy and robust performance across various environments, marking a notable achievement in usability and technological application in real-life scenarios. Additionally, feedback from users indicated a more than 90% success rate in real-time gesture recognition, emphasizing the system’s practical utility. This research not only bridges a vital communication gap for those who rely on BSL but also paves the way for future innovations, including integration into mobile and web platforms, thus expanding access and usability. The insights gained here highlight the profound impact machine learning can have on enhancing communication tools for the speech-impaired, resonating with ongoing trends in computational linguistics and interactive technologies