AI-Powered Speech Recognition System Developed to Boost Bambara Child Literacy

Automatic speech recognition (ASR) technology holds significant promise for improving literacy assessment, particularly for children. However, this potential remains largely untapped for many African languages, including Bambara, due to a lack of dedicated solutions. A new open-source initiative addresses this gap by presenting a comprehensive ASR system designed specifically for evaluating children's reading proficiency in Bambara.

The development process for this innovative system was end-to-end, beginning with extensive field data collection. Researchers utilized a mobile application to gather 55 hours of raw reading speech from 60 children, forming a public benchmark dataset for Bambara child-reading assessment. This dataset was then used to train and fine-tune ASR models, comparing a Bambara-adapted Fast-Conformer framework called Soloni with a compact convolutional architecture named QuartzNet.

Experimental results demonstrated the superior performance of the Soloni model, which significantly reduced the Word Error Rate (WER) from 0.42 to 0.22 and the Character Error Rate (CER) from 0.15 to 0.08, surpassing QuartzNet on the isolated benchmark. Further analysis revealed that while repeated readings offered substantial benefits for QuartzNet, Soloni saw only marginal improvements. Crucially, disaggregated data identified children under ten as the primary source of remaining errors, suggesting a need for more targeted data collection from younger readers in future iterations. The application's effectiveness was further validated through ten successful classroom trials.

This open-source ASR solution represents a vital step forward for literacy education in Bambara-speaking communities across Africa. By providing a reproducible and effective tool for assessing reading skills, it can empower educators and support children's learning journeys. The project not only offers a concrete application but also lays groundwork for further research and development in ASR for other underserved African languages, highlighting the potential of AI to address critical educational challenges on the continent.