| 方法 | PER (%) | WER (%) |
| Encoder-decoder LSTM [14] | 7.63 | 28.61 |
| Joint sequence model [13] | 5.88 | 24.53 |
| Joint maximum entropy (ME) n-gram model [18] | 5.90 | 24.70 |
| End-to-end CNN [19] | 5.84 | 29.74 |
| Encoder-decoder LSTM with attention [19] | 5.68 | 28.44 |
| Transformer 4 × 4 [5] | 5.23 | 22.10 |
| Transformer 4 × 4 | 5.46 | 21.91 |