Using Captions and Transcription to Enable Inclusive Learning
Vision and Values
The Ono Academic College has committed itself to creating meaningful change within Israeli society.” Guided by this vision, the Centre for Academic Support and Accessibility is advancing innovative solutions that promote equal participation for students with disabilities in higher education.
The Challenge: Deaf and Hard of Hearing Students in Academia
Deaf and hard of hearing students face significant barriers in academic environments, including difficulties accessing spoken lectures, reliance on human-mediated support, limited availability of captioned recordings, and inconsistent transcription quality. These challenges are particularly evident in content-heavy courses and discussion-based teaching.
Identifying the Need
The pilot initiative emerged in response to direct requests from students who require accurate and accessible captions, especially for recorded lectures, as a foundation for comprehension, revision and independent learning.
The Solution: A Hybrid Human–AI Approach to Captioning
The Centre for Academic Support and Accessibility at Ono Academic College has launched a pilot based on a hybrid model that combines AI-generated captions with human refinement. While AI enables fast and scalable access to lecture captions, human involvement ensures accuracy, clarity and sensitivity to the linguistic and academic complexity of Hebrew. This approach reflects a core value of accessibility work: technology serves inclusion most effectively when it is guided by human expertise, responsibility and commitment to equal participation for all students.
Benefiting Wider Student Populations
Beyond deaf and hard of hearing students, AI-generated captions can support students with attention deficit disorders and those studying in Hebrew as a second language, enhancing focus, comprehension and retention of academic content.
Technology as a Tool for Social Inclusion
The integration of AI into accessibility practices reflects not only technological innovation, but a broader social commitment: reasonable adjustments are a prerequisite for equal participation in higher education.
Implementation Challenges
Alongside its significant potential, the implementation of AI-based captioning presents several challenges:
Technical infrastructure
Research in speech recognition consistently demonstrates that audio quality is a key determinant of transcription accuracy. Microphone quality, lecturer proximity, and background noise levels all directly affect system performance.
Collaboration with academic staff
Effective optimisation requires close cooperation with lecturers, including awareness of clear pronunciation, appropriate speaking pace, consistent terminology, and minimising overlapping speech. Such practices substantially improve automatic captioning outcomes.
Linguistic challenges in Hebrew
Hebrew presents unique linguistic barriers for AI-based speech recognition, particularly in academic contexts. Its lack of written vowelisation, frequent homophones and rich morphology increase transcription error rates, especially when specialised terminology is used. As a low-resource language in speech processing, Hebrew requires systems to infer pronunciation and meaning from context alone, making automated captioning more sensitive to speaking style, audio quality and domain-specific language.
Current solutions and human refinement
To address these challenges, many systems adopt a human-in-the-loop approach, combining AI-generated captions with post-editing or quality control by trained human reviewers. Human refinement enables correction of linguistic ambiguities, specialised terminology and contextual errors, significantly improving accuracy and usability. While this hybrid model enhances quality, it also highlights the continued importance of professional expertise alongside technological innovation.
Looking Ahead
This pilot represents an initial step towards broader adoption of AI-based accessibility tools. By combining technological innovation with human expertise and institutional commitment, Ono Academic College continues to advance inclusive excellence and equal opportunities in academia.
References
Turetzky, A., Tal, O., Segal-Feldman, Y., Dissen, Y., Zeldes, E., Roth, A., Cohen, E., Shrem, Y., Chernyak, B. R., Seleznova, O., Keshet, J., & Adi, Y. (2024). HEBDB: A weakly supervised dataset for Hebrew speech processing. Retrieved from https://pages.cs.huji.ac.il/adiyoss-lab/HebDB/
Article meta data
Clicking on any of the links in this section will take you to other articles that have been tagged in the same category.
- This article was written by the ONO Karten Centre
- Featured in the Karten Winter 2026 Newsletter
- This article is listed in the following subject areas: Centre News, Other