Braille to Text and Voice Using Convolutional Neural Network

Abstract

Braille is an essential communication system for visually impaired individuals, but converting printed Braille into digital text and speech remains a major challenge. This paper presents an intelligent Braille Recognition System using Convolutional Neural Networks (CNN) for recognizing Braille characters from images and converting them into readable text and speech output. The proposed system uses MobileNetV2 architecture along with OpenCV-based image preprocessing techniques such as grayscale conversion, adaptive thresholding, segmentation, and noise reduction to improve recognition accuracy. The recognized Braille text is converted into speech using the pyttsx3 Text-to-Speech engine, enabling real-time audio output. The system provides a user-friendly web interface developed using Flask, HTML, CSS, and JavaScript for uploading images and viewing results. Experimental results demonstrate high recognition accuracy and efficient real-time performance. The proposed system improves accessibility, promotes inclusivity, and helps visually impaired individuals access printed Braille documents independently.

Introduction

This project presents an AI-based Braille Recognition System designed to convert Braille images into digital text and speech, improving accessibility for visually impaired users. Since manual Braille transcription is slow and requires specialized expertise, the system automates the process using deep learning and image processing techniques.

The system employs OpenCV for image preprocessing and the MobileNetV2 Convolutional Neural Network (CNN) for Braille character recognition. Recognized text is converted into speech using the pyttsx3 Text-to-Speech (TTS) engine, enabling users to access information independently.

The literature review highlights previous research on Braille-to-text and speech conversion, including CNN-based recognition systems, OCR-assisted translators, bidirectional Braille-speech communication tools, and image preprocessing methods for optical Braille recognition.

The project addresses challenges such as poor image quality, lighting variations, Braille dot misalignment, and the limitations of existing solutions that depend on expensive hardware or provide inaccurate outputs. Its main objectives are to recognize English Braille characters, preprocess images, generate readable text, produce speech output, and provide a user-friendly web application.

The methodology includes seven stages: image acquisition, image preprocessing, Braille segmentation, CNN-based character recognition, text generation, optional translation, and text-to-speech conversion. Users upload Braille images through a web interface, after which OpenCV techniques such as grayscale conversion, adaptive thresholding, Gaussian blur, and noise removal are applied. The segmented Braille cells are classified by the MobileNetV2 model, converted into English text, and then transformed into speech.

The system follows a client-server architecture consisting of five main modules: Input, Preprocessing, CNN Recognition, Text Processing, and Text-to-Speech. Together, these components provide an efficient and accessible solution for converting Braille documents into text and voice outputs.

Conclusion

The Braille to Text and Voice Conversion System successfully demonstrates the integration of deep learning, image processing, and speech technologies for improving accessibility. The system accurately recognizes Braille characters from images and converts them into text and speech output. The project promotes inclusivity and provides an effective assistive technology solution for visually impaired individuals. Future enhancements may include multilingual Braille support, mobile application integration, and real-time camera-based recognition.

References

[1] P. Bangar, S. Deshmukh, and A. Patil, “Braille-to-Text and Voice Conversion Using CNN,” International Journal of Research, vol. 9, no. 4, pp. 45–50, 2024. [2] M. Ansari, F. Shaikh, and R. Khan, “Braille Translator using CNN and OCR,” IEEE International Conference on AI Systems, pp. 112–118, 2024. [3] A. Pérez-Aguirre, I. González-Carrasco, and L. De Marcos, “Bidirectional Braille-Speech Communication,” Springer Journal of Assistive Technologies, vol. 15, no. 2, pp. 67–79, 2024. [4] A. Shirsekar, D. Patil, and S. Mahajan, “Image Preprocessing for Optical Braille Recognition,” International Journal of Computer Applications, vol. 162, no. 5, pp. 23–28, 2017. [5] Howard et al., “MobileNetV2: Inverted Residuals and Linear Bottlenecks,” IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018. [6] OpenCV Documentation. Available: https://opencv.org [7] Flask Documentation. Available: https://flask.palletsprojects.com [8] TensorFlow Documentation. Available: https://www.tensorflow.org

Copyright

Copyright © 2026 Akshara S Sundar, Jeevitha C J, Lisha S P, Vidyashree Y S, Mr. Sanjay M. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.