“2020-04-06”版本间的差异

来自cslt Wiki
跳转至: 导航搜索
 
(3位用户的3个中间修订版本未显示)
第91行: 第91行:
 
|Jiawen Kang
 
|Jiawen Kang
 
||  
 
||  
*  
+
* Features extractor code.
 +
* Calcuate EER.
 
||
 
||
*  
+
* Multi-genre scoring.
 
||
 
||
 
*   
 
*   
第114行: 第115行:
 
|Sitong Cheng
 
|Sitong Cheng
 
||  
 
||  
*  
+
* Get the result of model trained by wsj+ce+je
 +
* Get the result of model trained by vox+cnceleb
 
||
 
||
*  
+
* Do something about lid
 +
* More tests on model trained by vox+ce+je
 +
* Find the reason of result of EN&CN
 
||
 
||
 
*   
 
*   
第125行: 第129行:
 
|Zhixin Liu
 
|Zhixin Liu
 
||  
 
||  
*  
+
* Finish infrogan and gmm-ubm
 
||
 
||
*  
+
* Prepare baseline of language identification
 
||
 
||
 
*   
 
*   

2020年4月7日 (二) 00:16的最后版本

People This Week Next Week Task Tracking (DeadLine)
Dong Wang
  • More work on NPCA. Experiments show it works, but I still cannot find a reasonable probabilsitic framework to explain the results
  • Some work on NF-based on MNIST generation, with MSE loss. No good performance is obtained right now.
  • More investigation on NPCA. If cannot get result, skip it right now.
  • Investigation on conditional NDA.
  • Investigation on approximation power of NF.
Yunqi Cai
Zhiyuan Tang
  • Phone-ali conditioned NDA. Struct not work yet.
  • Continue.
Lantian Li
  • Overview Bayesian scoring methods.
  • Obtain a basic result of NDA scoring on x/i-vector.
  • Go on study NDA scoring.
Ying Shi
  • investigate sequential flow (supervise & unsupervise)
  • experiment about linear transfor of z
  • continue linear transfor of z
Haoran Sun
Yue Fan
  • Work on the arrangements of cn2
  • Experimented with gun and nongun on dnf
  • Do more experiments on dnf for classification
Jiawen Kang
  • Features extractor code.
  • Calcuate EER.
  • Multi-genre scoring.
Ruiqi Liu
  • Study meta-learning experiments and get the EER in different domains.
  • Arrange ASR training tools.
  • Continue meta-learning experiments.
Sitong Cheng
  • Get the result of model trained by wsj+ce+je
  • Get the result of model trained by vox+cnceleb
  • Do something about lid
  • More tests on model trained by vox+ce+je
  • Find the reason of result of EN&CN
Zhixin Liu
  • Finish infrogan and gmm-ubm
  • Prepare baseline of language identification
Haolin Chen
  • Some investigations on sequential flow
  • Cooperate with Shi Ying on experiments