In today's digital world, the trend of using eBooks is increasing day by day. Luckily, the digital world is also more accessible for persons with visual impairment. They use screen reading software like NVDA, Google TalkBack, etc. to interact with the digital world. One of the most common formats of eBooks is PDF. Some of them are digitally generated and other have scanned images of the printed book. PDF is one of the best formats for device independent visual rendering. Due to the lack of structural information, it doesn't provide appropriate accessibility to persons with visual impairment. It becomes more challenging when it comes to STEM ebooks. This project explores the tools and techniques required for meeting the above requirement.
A vast variety of mathematical content is available in the form of non-standard encoding and images. Tools and techniques are required to convert them into standard formats like MathML (Mathematics Markup Language). Even mathematical content in standard format has complexity in audio rendering due to its two-dimensional nature and contextual dependency.
The understanding of diagram is crucial for understanding STEM content. It reduces the amount of effort required to understand the critical concepts and provide better visualization. Sometime, even the textual description in the book is highly ambiguous without access to the diagram. Similarly, for graphs, its hard to get visualization from tabular data. Diagram based questions also enhance the reasoning ability of the student. Visually Impaired students are generally deprived of the access of the diagrams. This requires Alt-text generation/augmentation, which becomes more challenging when it comes to handling verbosity on the basis of context. Image description APIs are currently far behind in providing the adequate description of textbook diagrams. Tools and techniques are required for automatic alt-text generation/augmentation.
Indian Language Content Accessibility
Textual content available in regional languages is inaccessible due to use of non-unicode encodings. Screen readers are generally multi-lingual, that's why they require content to be in unicode fonts (UTF-8). This requires conversion scripts which can map non-unicode fonts to the unicode fonts. Not only this, generally publishers in India use Pagemaker or CorelDraw for book publishing. Both of these software are not compatible with unicode fonts. Conversion of these master files to format like Indesign with proper tags is also required.
Once we have eBook in the accessible format, a reader is required, which can support all the standards and provide appropriate navigation gestures. Here we are working on an ePub reader under the collaboration with VoxOMos Systems Pvt. Ltd.
- Human and Machine Speech Communication: Audio digit recognition using HTK
- Machine Learning: Document Classifier using Naive Bayes Algorithm.
- Special module on Intelligent Information Processing: Deep learning based hand-written digit recognition on NLTK handwritten digit dataset.
- Image Processing: Colourisation using histogram matching, Seam carving.
- AudioBook, Inclusive STEM Hackathon 2018 (Jan 2018)
A webapp for audio-book creation using crowd-sourcing. It is optimised for minimising for spam emails to the volunteers. This can be extended for various services like scribe, note-taker or any kind of service, which requires volunteering. It is developed using flask and MySQL.
- Emma: Disability bot, International Accessibility Summit, IIT Madras (Jan 2017)
A Quera type platform for disability related questions. Please check out its demo here. It was developed using Nodejs and MySQL.