“Zhiyuan Tang 2015-12-07”版本间的差异
来自cslt Wiki
| 第14行： | 第14行： | ||
3. something about Memory net. | 3. something about Memory net. | ||
| − | |||
| − | |||
| 第25行： | 第23行： | ||
:*CTC/nnet3/Kaldi | :*CTC/nnet3/Kaldi | ||
===== conditioning learning ===== | ===== conditioning learning ===== | ||
| − | * language vector into multiple layers -- | + | * language vector into multiple layers --Zhiyuan |
:* a Chinese paper | :* a Chinese paper | ||
| − | * speech rate into multiple layers -- | + | * speech rate into multiple layers --Zhiyuan |
:*verify the code for extra input(s) into DNN | :*verify the code for extra input(s) into DNN | ||
2015年12月7日 (一) 06:57的最后版本
Last week:
1. launched experiments on 1400h-Chinese with CTC/nnet3/Kaldi;
2. verified the code for conditioning learning of dnn using speech rate as extra input(s), results in [1];
This week:
1. more research about ctc/nnet3;
2. find the reason why language vector doesn't help;
3. something about Memory net.
TASKS
End-to-End
- monophone ASR --Zhiyuan
- MPE
- CTC/nnet3/Kaldi
conditioning learning
- language vector into multiple layers --Zhiyuan
- a Chinese paper
- speech rate into multiple layers --Zhiyuan
- verify the code for extra input(s) into DNN
