Bookbot

Bookbot is an Al-powered digital reading tutor designed to develop literacy skills through interactive reading support. With support from our grant and University of Dar es Salaam, the app was adapted to Kiswahili for Tanzania. It uses offline Al technologies, including Automatic Speech Recognition (ASR) to capture and transcribe children's reading and Text-to-Speech (TTS) to guide students through the story and correct pronunciation.

Why this is important
Tanzania has a clear need for early literacy support, especially for students with special needs as only 10% receive schooling. Voice activated reading support in low resource languages can offer meaningful assistance for early and struggling readers, especially if supported with contextually relevant materials.

Key Learnings

1. Cultural relevance requires human and AI collaboration - AI-translated books accelerated library creation, but without cultural localisation and phonics checks, quality slipped. Bringing in local experts was key to ensuring cultural and linguistic fit.
2. Design for real-world classrooms, not just labs - Using noise-robust processing was needed alongside training on clean lab audio to ensure the app could perform reliably on low-end devices in school settings.
3. Every language can require custom approaches - Although Kiswahili is phonetically transparent (words are generally pronounced as written), TTS training still required multiple recordings and tailored training data to account for complexities with tempo, cadence, and regional accents.
4. Human review ensured quality of TTS outputs - Model failures were high on some automated quantitative evaluations. Cross referencing with qualitative human checks was important for reliable judgement of audio quality.

Grant we provided: $59,800