MoDiCoL: A Modular Diagnostic Continual Learning Dataset for Robust Speech Recognition

In three linesMoDiCoL is a modular continual learning dataset for evaluating ASR robustness under real-world distribution shifts (accents, noise, recording conditions, speech impairments). Authors propose a real-world-inspired curriculum and evaluate three continual learning strategies to analyze how robustness develops, transfers, and is forgotten.

Summary generated by Claude — human-verified