SILENCY - SIGN LANGUAGE DECIPHER PROJECT

Silency helps visually hearing impaired people to communicate in real time with society and raise their expressional voice. 

The Isolated Sign language is one of the most accepted communication methods for people with hearing and speech impairments. Sign language has received incredible attention and advances over the years. On the other side if we opt for the human translation service it can cost a money (may be proved as expensive). To solve this, we need such kind of product that is both versatile and robust. In this project, which proposes a method with aim of breaking down the barrier between the deaf and the rest through the realization of a sign language recognition system or a decipher. 

This project demonstrates the recognition of various hand gestures in Sign Language using a deep learning-based architecture, i.e., CNN. The main challenge in developing a sign language decipherer is to design a good classifier that can classify input layers, dense layers and static gestures with high precision and as few errors as possible.  The method which I’ve applied is, the hand is first passed by a webcam frame which captures images, gestures and processed to input filter, and then it is processes with image classifier which is Mobileet, which predicts the class of hand gestures or movements. For the 26 letters of the alphabets, few numbers and gestures, and then processed this image to CNN algorithm, thus technique has a 96% accuracy rate. 

I've presented a sign language decipherer approach that uses the Convolutional Neural Network (CNN) trained on custom data collected from the webcam. Using custom hands images, the method trained CNNs to recognize numerals, alphabets, and other commonly used terms. 

The most difficult aspect of being a sign language decipherer is creating a good classifier that can categorize the input static gestures with high accuracy and as little error as feasible. The proposed system trained the classifier with various settings and achieved good accuracy. 

Deep Learning, Jupiter NB - Data Pandas Profiling, MobileNet, CNN Patterns, LSTM, Html, CSS, Figma.