Fingerspelling for Indian Sign Language using Swin Transformer

In this study, the Swin Transformer is used to predict Indian Sign Language (ISL) alphabets (A-Z) and is incorporated into an interactive Streamlit-based interface. The study intends to improve the accuracy and usability of ISL recognition systems. After being refined on a unique ISL dataset, the Swin Transformer achieved an 88% test accuracy. Using Google Translator, pyspellchecker and gTTS libraries, the system can correct formed words, translate words and synthesize voice in regional Indian languages like Hindi, Tamil, Gujarati and in international languages like Japanese, French and Spanish. The user-friendly interface makes it easy for users to fingerspell words and obtain corrected text and translated audio outputs along with signed videos of the words. Future improvements include the ability to process data in real-time.