DotSpeak：实时盲文识别系统让视障人士触达数字世界

DotSpeak is an AI-based braille recognition system developed by Xmanish8 and released on GitHub (https://github.com/Xmanish8/DotSpeak) on June 1, 2026. It uses YOLOv8 and MobileNetV3 to convert physical braille into digital text and voice in real time, aiming to help the 253 million visually impaired people globally access digital content more easily.

Globally, about 253 million people have visual impairments, and braille is their main way to get written information. However, most digital content is visual, creating a digital divide in education, healthcare, and employment. For example, visually impaired people struggle to read drug labels, public signs, or office documents independently, limiting their social participation. DotSpeak was created to bridge this gap between tactile braille and the digital world.

DotSpeak is an end-to-end system with a dual-model integration strategy:

1. Visual Recognition Layer: Uses OpenCV to capture and preprocess images (denoising, contrast enhancement, region cropping).
2. Dual Model Engine: YOLOv8-cls (fine-tuned on braille datasets, input size 64x64) for main classification; MobileNetV3 as a validator to cross-check results, reducing misreading in noisy environments (e.g., uneven light, worn braille).
3. Confidence Visualization: Provides animated confidence bar charts and Top-5 predictions, allowing users to verify results when confidence is low.

Features:

• Supports all 26 English letters.
• Real-time inference (43ms per prediction with GPU acceleration).
• Offline operation (all computations local, protecting privacy).
• Result export (save as image frames).
• Dual training modes (Python scripts/Jupyter Notebooks).

Applications:

• Education: Convert braille textbooks to digital text for screen readers.
• Healthcare: Recognize braille on drug labels for independent medication management.
• Public Facilities: Decode braille signs for navigation.
• Office: Digitize braille documents for independent work.

Quick Start:

1. Clone the repo and create a Conda environment with Python 3.10.
2. Install dependencies and download pre-trained model weights.
3. Run the inference script (sample images are provided).

Technology Stack: Python 3.10, YOLOv8 (Ultralytics), PyTorch (for MobileNetV3), OpenCV (image processing), Matplotlib (visualization), Conda/Jupyter (training environment).

DotSpeak is open-source, sharing code, model weights, training methods, and dataset structures to lower the barrier for assistive technology development. The developer transparently used AI tools like GitHub Copilot (code completion) and Claude/ChatGPT (documentation). Its vision is: 'Technology should be a bridge, not a barrier.'—helping visually impaired people access the digital world more equally.

DotSpeak demonstrates the potential of computer vision and deep learning in accessibility. It combines technical innovation with social inclusion. As technology evolves, we expect more such tools to create a more equal and inclusive digital society.