Table 3.

Performance of the benchmark and proposed models on clinical test sets in the clinical field in terms of the word error rate.

Methods	WER-CSa, %	WER-ENb, %	WER-ZHc, %
Character-based transfer learning	60.40	107.59	56.88
Syllable-based transfer learning	33.65	85.58	30.31
Azure cloud application programming interface	31.75	96.08	27.80
Character-based meta–transfer learning (proposed)	43.57	114.06	38.41
Syllable-based meta–transfer learning (proposed)	31.06	89.52	27.84

^aWER-CS: word error rate–code-switching.

^bWER-EN: word error rate–English.

^cWER-ZH: word error rate–Zhōngwén.