Signify: A Real-Time Sign to Text and Text to Sign Mobile Application for Dynamic Filipino Sign Language Translation using Transformer Architecture Deep Learning Model

This study presents Signify, a real-time, bidirectional mobile application for dynamic Filipino Sign Language (FSL) translation designed to bridge communication gaps between the Deaf and Hard of Hearing (DHH) community and hearing individuals.Utilizing Long Short-Term Memory (LSTM) and Transformer architectures, the system enables Sign-to-Text (S2T) and Text-to-Sign (T2S) functionalities.To improve model robustness, the researchers expanded the FSL-105 dataset by adding a "Directions" category and recording 80 additional videos per gesture, resulting in a total of 11,530 videos.For S2T recognition, hand landmarks were extracted via MediaPipe.Comparative analysis revealed that the Transformer model significantly outperformed the LSTM baseline, achieving a test accuracy of 98.73%.This was further improved to 99.60% through data augmentation techniques including Gaussian noise injection and temporal jitter.The T2S module utilizes a direct mapping approach to retrieve prerecorded FSL video segments validated by a certified interpreter for linguistic accuracy.Integrated into an Android application using TensorFlow Lite, the system supports realtime, offline inference.User usability testing yielded a Grand Overall Mean of 4.57 (Excellent), reflecting high satisfaction among signers and non-signers.This research advances inclusive communication in alignment with SDGs 4 and 10.