Braille AI: a convolutional neural network-based mobile application for bidirectional Braille translation in Spanish

Students with visual impairments face significant educational barriers due to the limited availability of Braille-trained educators and accessible instructional materials. This study presents Braille AI, a native Android application designed to support bidirectional translation between Spanish text and Braille. Materials and methods: The application integrates two artificial intelligence models: YOLOv8 for Braille symbol detection and a convolutional neural network implemented with TensorFlow Lite for character classification. Public datasets, including Angelina, DSBI, and a Braille dataset from Kaggle, were used for model training. The system also generates visual guides for operating a Perkins Brailler machine to support educators with limited Braille knowledge. Results: Functional evaluations were conducted with two teachers from the USAER 68 support unit. The symbol recognition model achieved an average accuracy of approximately 97%, while translation processing was completed in under five seconds. Participants reported improved accessibility, reduced preparation time, and greater confidence in supporting students with visual impairments.

Braille AI offers a portable and cost-effective tool for improving communication between sighted educators and blind students. The system demonstrates the potential of artificial intelligence to enhance inclusive education in Spanish-speaking contexts.