“Mengyuan Zhao 2016-05-16”版本间的差异
来自cslt Wiki
(以“* sino 8k training: # Finish 1000h mpe training for the model with noise phone. results looks ok (significant deletion error decrease, and similar WER compared with...”为内容创建页面) |
|||
第1行: | 第1行: | ||
* sino 8k training: | * sino 8k training: | ||
− | # Finish 1000h mpe training for the model with noise phone. | + | # Finish 1000h mpe training for the model with noise phone. result looks ok (significant deletion error decrease, and similar WER compared with old model). |
# Start adaptation using PingAn-All data. | # Start adaptation using PingAn-All data. | ||
* 16k embedded model training: | * 16k embedded model training: | ||
# 5-500 mpe model training done (without svd), delivered to sino and to be tested. | # 5-500 mpe model training done (without svd), delivered to sino and to be tested. | ||
# 5-500 mpe model with svd, on training. | # 5-500 mpe model with svd, on training. |
2016年5月15日 (日) 16:38的最后版本
- sino 8k training:
- Finish 1000h mpe training for the model with noise phone. result looks ok (significant deletion error decrease, and similar WER compared with old model).
- Start adaptation using PingAn-All data.
- 16k embedded model training:
- 5-500 mpe model training done (without svd), delivered to sino and to be tested.
- 5-500 mpe model with svd, on training.