MoDiCoL: A Modular Diagnostic Continual Learning Dataset for Robust Speech Recognition
The MoDiCoL dataset systematically tests Automatic Speech Recognition (ASR) model robustness by varying linguistic content, speaker characteristics, and acoustic environments. Researchers introduce a continual learning curriculum simulating incremental changes in these factors and evaluate three learning strategies—Experience Replay, Representation-level Regularization, and Orthogonal Gradient Descent—to understand how models acquire and maintain robustness against real-world distribution shifts.