Progressive Refinement: An Iterative Pseudo-Labeling Approach for Mandarin-English Code-Switching ASR

Researchers introduced an iterative pseudo-labeling method to improve Mandarin-English code-switching automatic speech recognition (ASR). This approach leverages unlabeled data through pseudo-label generation and two-stage bilingual model training, significantly reducing error rates on SEAME datasets.