Low-resource Language Discrimination Towards Chinese Dialects with Transfer learning and Data Augmentation

In three linesCDDTLDA framework for Chinese dialect discrimination with scarce annotation resources. Uses transfer learning on ASR models, data augmentation (speed, pitch, noise), and self-attention to capture shared semantic features. Outperforms state-of-the-art on two benchmark corpora.

Summary generated by Claude — human-verified