Avanceña, Angelo SJ. <br/><br/>
E-salin:   development of a tagalog fsl-to-text and speech-to-text- and-fsl images in real-time web-based filipino sign language and neural machine translation systems for selected 3 major philippine languages: cebuano, ilocano, and waray/ 
Angelo SJ. Avanceña, Rose Belle G. Bolor, Sophia B. Espiritu, Annabela T. Ignacio, Eruel Andrei O. Marasigan, and Aaron B. Mendoza.-- 
 - Manila:	  Technological University of the Philippines,	  2025. 
 - xvii, 272pages:  29cm. 
<br/><br/> Bachelor's thesis<br/> 
<br/><br/>College Of Engineering.--<br/> 
<br/><br/>Includes bibliographic references and index. 
<br/><br/> e-SALIN is a real-time, web-based translation system designed to bridge the<br/>communication gap between the deaf and hearing communities in the Philippines. While<br/>many deaf individuals primarily use Filipino Sign Language (FSL), limited understanding<br/>among the hearing often leads to miscommunication and exclusion. e-SALIN addresses<br/>this by translating FSL into Tagalog text and into three widely spoken Philippine languages:<br/>Cebuano, Ilocano, and Waray. It promotes inclusive and respectful communication through<br/>gender-fair and culturally sensitive language. Powered by machine learning and computer<br/>vision; including CNN and LSTM models. The system accurately recognizes and translates<br/>signs in real time, achieving over 94% accuracy within three seconds. It also uses neural<br/>machine translation (NMT) models like marianMT and mBART, trained on a dataset of<br/>507 Tagalog words translated into regional languages. Despite working with low-resource<br/>languages, it maintains high translation quality, with BLEU scores ranging from 86.9% to<br/>98.8%, highlighting e-SALIN’s strong potential to support inclusive and multilingual<br/>communication. A dataset of 507 Tagalog words translated into Cebuano, Ilocano, and<br/>Waray using Google Colab Pro + was used to train the marianMT and mBART models.<br/>BLEU ratings showed that even with low-resource languages, translation quality varied<br/>from 86.9% to 98.8%, indicating good accuracy. These results highlight the system's strong<br/>multilingual translation and sign recognition capabilities. 
<br/><br/><label>Subjects--Topical Terms: </label><br/>Real-time system<br/>Web-based system<br/>Translation system
<br/><br/><label>LC Class. No.: </label>BTH TK 870  / A93 2025