“2020-02-24”版本间的差异

来自cslt Wiki
跳转至: 导航搜索
 
(5位用户的5个中间修订版本未显示)
第33行: 第33行:
 
|Zhiyuan Tang
 
|Zhiyuan Tang
 
||  
 
||  
*  
+
* Analyse  the structure of glow for disentanglement.
 
||
 
||
*  
+
* Analyse more steady noise for BP denoising.
 
||
 
||
 
*   
 
*   
第44行: 第44行:
 
|Lantian Li
 
|Lantian Li
 
||  
 
||  
*  
+
* Update frame-based DNF repository.
 +
** Modify class_mean initialization.
 +
** Update Container for class mean update.
 +
** Update visualization tools.
 
||
 
||
*  
+
* Start subspace DNF.
 
||
 
||
 
*   
 
*   
第55行: 第58行:
 
|Ying Shi
 
|Ying Shi
 
||  
 
||  
*
+
* DAE baseline
 +
* compare DAE and Double Flow on spectrum level
 
||
 
||
*
+
* compare DAE and Double Flow on SDR level
 +
* semi-supervise Double Flow
 +
* verification the denoise ability on unsteady noise of Double Flow
 
||
 
||
 
*   
 
*   
第66行: 第72行:
 
|Wenqiang Du
 
|Wenqiang Du
 
||  
 
||  
*
+
* Verify the differences in model performance between audio recordings on different devices
 
||
 
||
 
*
 
*
第88行: 第94行:
 
|Yue Fan
 
|Yue Fan
 
||  
 
||  
*
+
*Baseline line of svm
 +
*Data download
 
||
 
||
*
+
*Baseline line of svm&gmm
 +
*Experiments on dnf
 
||
 
||
 
*   
 
*   

2020年2月24日 (一) 00:57的最后版本

People This Week Next Week Task Tracking (DeadLine)
Dong Wang
  • Investigate between-class regularization, experiemnted with moment match
  • Investigate cosine and l2 distance, re-visit the l2-based scoring.
  • More investigation on l2 scoring
  • Contiue draft on DNF properties.
Yunqi Cai
  • Writing the patent of DNF
  • Test new code of DNF
  • Investigate the total covariance constrain of DNF
  • Finish the patent today
  • Process the experiment data of total covariance constrain
Zhiyuan Tang
  • Analyse the structure of glow for disentanglement.
  • Analyse more steady noise for BP denoising.
Lantian Li
  • Update frame-based DNF repository.
    • Modify class_mean initialization.
    • Update Container for class mean update.
    • Update visualization tools.
  • Start subspace DNF.
Ying Shi
  • DAE baseline
  • compare DAE and Double Flow on spectrum level
  • compare DAE and Double Flow on SDR level
  • semi-supervise Double Flow
  • verification the denoise ability on unsteady noise of Double Flow
Wenqiang Du
  • Verify the differences in model performance between audio recordings on different devices
Haoran Sun
Yue Fan
  • Baseline line of svm
  • Data download
  • Baseline line of svm&gmm
  • Experiments on dnf
Jiawen Kang
  • Make scripts to calculate subset EER;
  • Continiue model experiments;
  • Paper introduction part.
  • Length and age EER;
  • Continiue model experiments;
Ruiqi Liu
  • Continue training the model.
  • Experiments on CN-Celeb different scenes.
  • Experiments on CN-Celeb different genders.
  • Go on the task1.
  • Experiments on CN-Celeb different length.
  • Prepare paper writing.
Sitong Cheng
Zhixin Liu
Haolin Chen
  • Try Double Flow(Glow) with spectrogram as input, conditional/without condition.
  • Evaluation.
  • Optimize model structure.