Deep Learning based Bangla Voice to Braille Character Conversion System

Braille is a communication method for sight-disabled people, which has not been incorporated that much into the modern education system. Lots of research have been done on braille systems but most of them are about the conversion of a text character to a braille pattern. However, so far no work has been done with Bangla voice to braille character conversion. This paper introduces a system that will help visually impaired people to start their very basic education with a technology that is capable of recognizing Bangla characters. When a Bangla character is uttered by a person, our system will convert that character into a braille pattern. Afterward, a person using a finger touch can easily recognize the corresponding braille for that character. This voice recognition is done using the Visual Geometry Group (VGG-16) model of Convolutional Neural Network (CNN) and Arduino Uno. A 16x4 LCD display is used to generate the braille pattern for the recognized character. After training the model with 50 epochs, we achieved 98% of train accuracy and 92.42% of test accuracy from the test data.