P. Raghunadh

Abstract Low-resource speech recognition remains fundamentally limited by training distribution, as neural networks cannot access linguistic knowledge beyond their training data. SAREC (Semantic-Aware Retrieval Enhanced Conformer), a novel…