Hasta-Shabd: sign language translator

Sign Language is the only and important mode of communication for the normal humans to communicate with the deaf/mute community. Despite this there is a huge gap for communication between the sign language users and rest of the population. Hasta-Shabd is an extensive system which converts sign to text and audio to sign on the basis of Indian Sign Language. The project was developed in a total of 6 different stages with an aim to develop a Two-way system which has high accuracy. The system utilizes various CNN/RNN models(AlexNet, VGG19, Conv 1D, LSTM, ResNet 1D) to achieve this goal. On an average the train, validation and test accuracy was around and above 90%, while the train, validation and test loss was aroun d and lower than 0.7. The dataset which was used for the sign to text was taken from Kaggle while the dataset which was used for audio to sign was self-made. The dataset was augmented to make it versatile. The system was developed on the VS Code platform using the deep learning frameworks for training and optimization. The outcome is a scalable sign language translator system which uses sophisticated structures like CNN/RNN for providing viable solution.