Over 63 million people in India, or about 6.3% of the population, are hard of hearing. This group is expected to grow due to factors such as aging. Though they do not regard their hearing loss as a disability, but rather a different way of life, their social interactions can be significantly limited.

Our project focuses on creating a robust, scalable system to predict, interpret, and translate Indian Sign Language (ISL) in real-time, without requiring specialized hardware. ISL combines actions, facial expressions, and body language, differing from other sign languages like American Sign Language (ASL), which may use single-hand gestures; ISL usually uses both hands. This complexity presents challenges in developing an accurate machine learning model for ISL interpretation.

We annotated images by capturing them from various angles, about 23k in total and then augmented the data to create a more generalized model. This solution operates entirely without a backend, making it lightweight and highly efficient.