Michael Leventhal

Soloni, a Bambara-adapted ASR model, achieves a remarkable reduction in word error rate from 0.42 to 0.22, outperforming existing architectures and paving the way for effective literacy assessments in underrepresented languages.