Real-Time Web-Based Indonesian Sign Language (BISINDO) Translator System Using CNN-LSTM Deep Learning and Text-to-Speech

Imjauzanansyah, Nabiel Muhammad
Simbolon, Hevlie Winda Nazry

Jurnal ICT : Information and Communication Technologies
Vol. 17 No. 1 (2026): April, Jurnal ICT : Information and Communication Technologies

Indonesian Sign Language (BISINDO) is the primary communication medium for the deaf community, yet low public understanding often causes communication barriers. Previous sign language recognition studies mostly operated offline, lacked real-time web integration, and only produced text output. This study designs and develops a real-time web-based BISINDO translator system using a hybrid Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) method, integrated with a Text-to-Speech (TTS) feature. The dataset consisted of primary video data from 3 subjects, covering 11 category classes with 1,000 frames per class in grayscale format (100x89 pixels). The hybrid CNN-LSTM model was integrated into a React.js and Node.js web application (NzSignify). Testing results demonstrate that the model achieved 96% static accuracy based on Confusion Matrix evaluation. In real-time functional testing, an 80% Confidence Threshold effectively filtered incorrect gestures, enabling accurate translation of valid sign gestures into text and voice output.