Language Identification

MAP-Mix: Training-dynamics-guided data augmentation for spoken language identification, achieving ~2% F1 improvement over random mixup at ICASSP 2023.