|
|
| 第31行： |
第31行： |
| | |- | | |- |
| | |Yang Wei || | | |Yang Wei || |
| − | * Learn RNN language model training principle and tool. | + | * Crawl 530M universal corpus from zhihu.com(crawler list: https://docs.google.com/spreadsheets/d/1VNtgnnRiez8HjjhpwDXqBvewQZp-tkivOUVA5qk0rPg/edit#gid=0) |
| | || | | || |
| − | * Read the source code of RNNLM tool | + | * Read source code of LM rescore |
| | || | | || |
| | | | |
| People |
Last Week |
This Week |
Next Week |
Task Tracing(DeadLine)
|
| Mengyuan Zhao |
- Prepare 3rd period delivery.
- Clean up code, finished.
- Cross-compilation.
|
- Try to add nnet3.
|
|
- confirm general testset (20170512)
- Add nnet3 (20170622)
|
| Zhiyong Zhang |
- Finished the run-pass for the phone-based TTS, but poor performance.
|
- To remove the silence from linguistic feature extraction
|
|
- To finish the phoneme-based KALDI-TTS</font>(20170616,20170623);
|
| Yang Wei |
|
- Read source code of LM rescore
|
|
- Similar-pair on Tonghuashun's LM.
|
| Dong Wang |
- Some discussion about the partner ship
|
- Start the first partnership project
|
|
|
| Mingrui Ruan |
|
- 官网建设的商讨
- 同花顺的项目节点及推进
- 与爱维尔合作的协议
- 拜访一个客户
|
|
|
| Zhenlong Han |
- Traing both uni and bi CTC model on data 1000h.
- 1000h ctc-uni model has been trained to complete.
|
- Modify dnn output pdf. but it does not work, I check the code many times, it's seems like the code is no bugs.
|
|
- Test various output level of CTC
- LSTM layer number (20170515)
- Frame skipping (20170515)
- First version of "wake up".
|
