Sometin Beta Pass Notin (SBPN): Improving Multilingual ASR for Nigerian Languages via Knowledge Distillation

In three linesNew multilingual ASR approach for Nigerian languages using knowledge distillation. SBPN model (120M and 600M parameters) covers Yoruba, Hausa, Igbo, Nigerian Pidgin, and Nigerian English. Achieves 29% average WER reduction vs monolingual baselines, outperforms SOTA models on Common Voice and Fleurs benchmarks.

Summary generated by Claude — human-verified