Braille to Speech conversion using BT-CNN

---

Abstract

Braille is a universally accepted system of writing and understanding scripts for blind people. It is a touch-based system for reading and writing. The alphabet is composed of specific dot configurations. An image containing a word or statement in Braille can be given to the model. Although there are various ways to implement it, the proposed model uses Convolutional Neural Network (CNN) to analyze the dots in the Braille image and convert them into letters, eventually forming words and statements. Keras library is to perform several operations like generating image data, splitting the dataset and evaluating the model's accuracy. This text is further processed by the model and the final output is an audio file which is a speech of the text obtained from the Braille Image. This text to speech conversion is done by using Google Text to Speech. By training the model with a large number of epochs, an accuracy of about 96.30% is achieved.

---

Keywords

Convolutional Neural Network; Deep Learning; Machine Learning; Speech Synthesis