PENGEMBANGAN SISTEM PENERJEMAH BAHASA ISYARAT INDONESIA KE DALAM TEKS SECARA REAL-TIME UNTUK KOSAKATA TERBATAS MENGGUNAKAN MEDIAPIPE DAN LONG SHORT-TERM MEMORY (LSTM)

Mimo Sukma Aliefasa (Dian Nuswantoro University)
Arry Maulana Syarif (Dian Nuswantoro University)

RABIT: Jurnal Teknologi dan Sistem Informasi Univrab
Vol 11 No 2 (2026): Juli

In an increasingly connected society, the ability to understand each other is key to bridging differences. For the deaf community in Indonesia, Indonesian Sign Language (BISINDO) is the primary bridge for expressing their thoughts and feelings. However, when this language encounters a public environment where the majority do not understand it, communication often stalls. This situation inspired this research, which developed a system capable of recognizing and converting BISINDO gestures directly into text. The classification process uses the Long Short-Term Memory (LSTM) method, while hand, face, and body frame detection utilizes MediaPipe Holistic This study utilizes a dataset of 110 videos involving 10 participants to perform 10 BISINDO vocabularies frequently used in everyday conversation and 1 idle gesture class. Real-time testing results show that a model with 11 classes, using a bidirectional LSTM architecture, 1000 epochs, 64 hidden layer units, and a batch size of 32, is able to achieve 91.25% validation accuracy.