Developing and adapting voice AI for inclusive early grade reading assessments: insights from Sub-Saharan Africa and South Asia

Group Submission Type: Formal Panel Session

Proposal

Early reading assessments play a key role in improving outcomes in foundational learning, however many children lack access to timely, high-quality assessments which therefore inform subsequent teaching practice (UNESCO & African Union, 2023). Without accurate data on learning, systems face challenges in identifying and supporting struggling learners, thus leaving gaps that can widen social divides and leave marginalized learners even further behind (World Bank, 2023). Advances in voice-based Artificial Intelligence (AI), particularly Automatic Speech Recognition (ASR), are a promising way to bridge these gaps by making assessments more frequent, inclusive, and context-sensitive to support more targeted and responsive teaching for improved learning outcomes.

This panel explores innovations in ASR for early grade reading, focusing on the ethical, technical, and operational challenges of developing tools in multilingual, low-resource contexts. Drawing on experiences from Tanzania, Kenya and India, presenters will share how child speech data are collected responsibly, how ASR models are adapted to challenging classroom environments, and how tools are integrated into existing education platforms to support teachers and learners. By centering inclusion and equity whilst scaling, the panel highlights how technology can strengthen learning systems.

Key Themes
Collectively, the panel will address the following key themes:

Assessment of early reading in multilingual settings: The panel examines how ASR can improve early reading assessments across diverse linguistic contexts, ensuring that local languages are appropriately designed and incorporated. Presenters will share results from developing Kiswahili, HIndi and Marathi models and multilingual applications trained on extensive child speech datasets, optimized for children's voices, classroom acoustics, and the most technically demanding reading subtasks including letter-sound identification, syllable reading, and oral reading fluency.

Ethics, child rights and legal compliance: Collecting and processing child speech data raises significant ethical and legal considerations that can limit AI innovation in low- and middle-income countries. The panel will discuss navigating informed consent processes, data protection protocols, and compliance with national and international standards, sharing practical lessons from field implementation that enable innovation.

Diagnostic precision: Presenters will demonstrate how AI-enabled assessments can uncover hidden learning gaps and create individualized learner profiles that provide actionable insights for teachers and learners. The tools are trained for granular error analysis - capturing omissions, substitutions, insertions, hesitations, and mispronunciations to generate individualized learning profiles. This includes distinguishing between decoding challenges, phonemic awareness gaps, and comprehension issues, while maintaining functionality in offline, low-connectivity classroom environments.

Systems integration: Presenters will explore how to enable the systematic adoption of early literacy assessment in low- and middle-income country contexts and facilitate data-informed, scalable student and teacher support, thus helping to improve student learning outcomes. They will share how this can be achieved through integration with established platforms, teacher dashboard development, and real-time feedback systems that reduce assessment burden while providing actionable insights at student, classroom, and system levels. Drawing on rigorous development processes including user testing, field testing, and pilot implementation with hundreds of students across multiple grades, the panel will discuss how technical design, local partnership development, and iterative refinement enable responsible scaling while maintaining diagnostic accuracy and cultural responsiveness.

The session supports the CIES 2026 theme, “Re-examining Education and Peace in a Divided World”, arguing that assessment technology can be a driver of justice and cohesion through targeting the intersections of the most marginalized learners. When effectively designed, these innovations make visible the learning needs of marginalized students, enabling targeted support and helping education systems close divides and supports greater equity.

Sub Unit

• General Pool

Chair

• Alice Cornish, Better Purpose

Individual Presentations

• Digitization of early grade reading assessments in Tanzania using voice AI - Romana Kropilova, Fab Inc

• Findings from developing an AI-supported early literacy assessment for student self-assessment in Tanzania - Carmen Strigel, RTI International

• Leveraging AI-based reading assessment to bridge foundational literacy gaps - Nishant Baghel; Vani Shree, Pratham International

Discussant

• Clio Dintilhac, BMGF