Researchers have developed a new multilingual Automatic Speech Recognition (ASR) framework called Sometin Beta Pass Notin (SBPN) to improve performance for Nigerian languages. The framework uses a two-stage knowledge distillation process, first from monolingual models and then through iterative self-improvement with pseudo-labeled data. This approach achieved an average 29% relative Word Error Rate reduction over baselines and outperforms existing state-of-the-art multilingual models on benchmarks like Common Voice and Fleurs. SBPN is released as open foundation models in two sizes, aiming to provide crucial ASR resources for the region. AI
Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →
IMPACT Provides open-source ASR models for under-resourced Nigerian languages, potentially enabling new applications and research.
RANK_REASON Publication of an academic paper detailing a new model and methodology. [lever_c_demoted from research: ic=1 ai=1.0]
