Table 4.
Model | Recognition Accuracy(%) | |||||
---|---|---|---|---|---|---|
Chinese Test Data | Tibetan Test Data | |||||
Mandarin | Cantonese | Hokkien | Ü-Tsang | Amdo | Kham | |
Hard-MTL Transformer | 99.73 | 99.19 | 98.21 | 99.38 | 99.58 | 99.32 |
Soft-MTL Transformer | 99.93 | 99.71 | 97.87 | 99.73 | 99.86 | 96.62 |
Adaptive Soft-MTL Transformer | 99.73 | 99.34 | 97.31 | 99.85 | 99.91 | 100.00 |