A Filipino Sign Language (FSL) Software: Conversion of FSL-to-Text and Speech Using Deep Learning

Abstract

Sign language has become a crucial instrument for impaired people to communicate with others. However, the lack of knowledge and mastery regarding sign language became a major impediment in society, creating a communication barrier between impaired and non-impaired people. With this concept of a sign language translator, the observed language barrier for deaf-and-mute and ordinary people was addressed. The software provided translation corresponding to the presented hand gestures in front of the camera or monitor of the system. This study used two design methods: experimental and developmental approaches. The first phase in creating the project was planning the features, including understanding Filipino Sign Language (FSL) fundamentals. Then, programming module installation and code debugging were put into practice.

To ascertain the accuracy of the system in terms of sign and range, five gestures were tested. According to the testing results, the translation for each hand gesture/sign language was determined to be accurate. The average accuracy for each sign and the cumulative average for all signs were calculated and solved using statistical data analysis. The results in sign accuracy demonstrate that all gestures were successfully translated. The study also identified the different range accuracy: Range Accuracy (1ft), Range Accuracy (2 ft), Range Accuracy (3ft), and Range Accuracy (4ft). The range accuracy at a distance of 1 foot was accurately translated.

Similarly, the range accuracy at 2 feet is also perfectly translated. The system can scan signals from two feet away with pinpoint accuracy. However, in the range accuracy at a distance of 3 feet, only 98% of the ranges are accurate. With this result, the system could still translate the gestures almost perfectly. The distance of 3 feet is far from the laptop, but it can still translate FSL gestures.

Furthermore, with the range accuracy at a distance of 4 feet, the system translated every sign with an accuracy of 86%. This received the lowest percentage of range accuracy because it is explicitly stated that the percentage result decreases as your distance from the system increases. The system was able to convert FSL into text and voice. The range of the signer to the system affects its translation. On the other side, the system sign accuracy was perfectly translated.