“14-30-10 Zhiyong Zhang”版本间的差异

来自cslt Wiki
跳转至: 导航搜索
(以内容“Weekly Report: 1, Verify the dither option effect on the wav feature extract. 2, Finished the WSJ wav2amr2wav's increment cross-entropy training based original ...”创建新页面)
 
 
第6行: 第6行:
 
     1, Continue the gmm-dnn co-training
 
     1, Continue the gmm-dnn co-training
 
     2, Sinovoice 6000h mpe training
 
     2, Sinovoice 6000h mpe training
     3, Using 1700h mpe 4-5.nnet do 6000h alis and denlats
+
     3, Using 1700h mpe 4-5.nnet to do 6000h alis and denlats

2014年3月11日 (二) 01:47的最后版本

Weekly Report:

   1, Verify the dither option effect on the wav feature extract.
   2, Finished the WSJ wav2amr2wav's increment cross-entropy training based original wav DNN model.

Target of next week:

   1, Continue the gmm-dnn co-training
   2, Sinovoice 6000h mpe training
   3, Using 1700h mpe 4-5.nnet to do 6000h alis and denlats