Adaptive Fisher-Whitened Cross-Covariance for Low-Resource ASR

FCCA-based adaptations for Whisper and Qwen3-ASR improve low-resource speech recognition by using task-informed parameter-efficient fine-tuning. Extensions like AC-FCCA and AR-FCCA allow for better reallocation of adaptation capacity across projection matrices under fixed budgets.