Explainable and Context-Aware Speech-to-Text for Special and Inclusive Education: A Systematic Review and a Proposed Conceptual Framework

Abstract

Background: Speech-to-text (STT) systems built on automatic speech recognition (ASR) are widely used to caption lectures and classroom talk and to support learners who are deaf or hard of hearing, who have communication or language-processing differences, or who learn in multilingual settings. Recognition accuracy has improved markedly, yet errors persist in authentic conditions and are typically delivered without any indication of which words are unreliable or why.

Objective: This review synthesises evidence on the reliability, transparency, and explainability of STT used as an access technology in special and inclusive education, and identifies the gap that motivates a context-aware, explainable post-recognition approach.

Method: Following the PRISMA 2020 statement, peer-reviewed, English-language records (January 2009–May 2026) were sought in Scopus, Web of Science, ERIC, IEEE Xplore, and PubMed, with citation and hand searching. Records were screened against pre-specified eligibility criteria; included studies were charted and synthesised narratively across four themes.

Result: The synthesised evidence shows that (a) STT meaningfully improves access for learners with disabilities but its usefulness is bounded by accuracy and reliability; (b) recognition degrades for atypical speech, child speech, and non-mainstream dialects, raising equity concerns; (c) confidence and post-recognition correction can flag errors but are rarely exposed to users; and (d) explainability—though central to trustworthy educational AI—has scarcely been applied to STT. No identified system combined reliability awareness with user-facing explanation in an accessibility context.

Conclusions: The review reveals a transparency gap and derives a conceptual framework, ECASTT (Explainable Context-Aware Speech-to-Text Technology), that augments any ASR engine with multi-signal reliability assessment and human-readable explanation. The framework is presented as a research agenda, not an evaluated system, and is aligned with Universal Design for Learning. Implications for assistive-technology design and a programme of user-centred validation are outlined..