“Zhiyong Zhang 2016-04-18”版本间的差异
来自cslt Wiki
| 第1行： | 第1行： | ||
*Weekly report | *Weekly report | ||
| + | # Re-compile nnet3-mpe under the 20160410 kaldi version | ||
| + | # Assist the restaurant recording of Siason | ||
| + | # Assist the implementation of NG-SGD in nnet1 version with Xiangyu | ||
# Character-LM and Word-LM merge | # Character-LM and Word-LM merge | ||
:# Union the two can achieve 9.53-wer compared with only word-LM 9.67-wer, but the HCLG is twice bigger than only word-lm; | :# Union the two can achieve 9.53-wer compared with only word-LM 9.67-wer, but the HCLG is twice bigger than only word-lm; | ||
:# Compose get an wer of 64-wer, need deep investigation | :# Compose get an wer of 64-wer, need deep investigation | ||
| − | |||
| − | |||
| − | |||
| − | |||
* Next week | * Next week | ||
# To solve the deletion error in discriminative training | # To solve the deletion error in discriminative training | ||
# To check the SVD re-train failure problem | # To check the SVD re-train failure problem | ||
2016年4月18日 (一) 00:53的最后版本
- Weekly report
- Re-compile nnet3-mpe under the 20160410 kaldi version
- Assist the restaurant recording of Siason
- Assist the implementation of NG-SGD in nnet1 version with Xiangyu
- Character-LM and Word-LM merge
- Union the two can achieve 9.53-wer compared with only word-LM 9.67-wer, but the HCLG is twice bigger than only word-lm;
- Compose get an wer of 64-wer, need deep investigation
- Next week
- To solve the deletion error in discriminative training
- To check the SVD re-train failure problem
