Projektbeschreibung
In June 2025, the EU Accessibility Act and the corresponding national Accessibility Improvement Act came into force. They require service providers to make products such as websites and e-books accessible to people with disabilities, e.g. visual impairments. Some exceptions apply, but, in practice, there is no full consensus yet which accessibility measures are sufficient to offer a good learning process for students with disabilities. One prominent aspect of accessibility is the presentation of contents in a way that they can be processed by assistive technology such as speech synthesis.
In fact, learning by listening does not only benefit students with disabilities, but is common among students of all cohorts. The IU Learning Report 2024 has established the broad usage of media such as podcasts and audio books among all age groups.
While the speech synthesis of plain texts has reached highly sophisticated levels for many years, the technology for reading mathematical formulas is still in its infancy. Though pipelines that convert text with formulas to audio-transcription exist in isolation, their integration has been barely tested and, more importantly, the quality of the resulting experience has not been measured. The project Read-Aloud Maths aims to investigate challenges and solutions for all realistic learning situations where written texts and formulas are “read-aloud” by means of speech synthesis.
More precisely we want to answer:
Q1: How can the accessibility of mathematical content in digital form be assessed from the perspective of real users employing standard and advanced assistive technologies (at IU and beyond)?
Q2: Which types of errors and ambiguities most significantly affect usability and learning outcomes, and how can they be systematically detected and mitigated in the authoring workflow?
Q3: How do different learner groups – distinguished by disability, learning context, and user preference – interact with and benefit from mathematical speech output, and what are their specific requirements for language, interactivity, and reliability?
To answer these questions, this incubator will select representative course content and enrich it under the eyes of professional mathematicians, who will ensure proper semantics. It will experimentally measure learning processes using spoken mathematical content. Alongside it will generate a ground truth corpus of expressions and their audio speech, in English and German. The enrichment will be so complete that one can obtain a realistic representation of the future conversion processes. Both the tools for the enrichment and the tools for reading-aloud will include widespread accessibility tools (e.g. web-browsers, rule-based accessibility tools) as well as state-of-the-art large language models’ pipelines to compare and to convert content.
04/2026 – 03/2028