“2019-12-09”版本间的差异

来自cslt Wiki
跳转至： 导航、 搜索
第102行： 第102行：
 
|Jiawen Kang
 
|Jiawen Kang
 
||  
 
||  
−
*  
+
* Sent materials to CUHK;
 +
* Reading paper about model robustness;
 +
* Kaldi baseline checking.
 
||
 
||
−
*  
+
* Experiences about CN-Celeb;
 +
* Preparing for CUHK inverview.
 
||
 
||
 
*   
 
*   

2019年12月9日 (一) 00:11的版本

People This Week Next Week Task Tracking (DeadLine)
Dong Wang
  • MelGan and WaveGlow re-investigation. Good neural vocoder is done.
  • Paper review on ICASSP 19, Interspeech 19.
  • Investigate the performance degradation with pure VP-based x-vector regularization compared to the VP+linear model
  • Design more concrete way of 'supervised flow', that involves supervision in sparse dimensions.
  • Paper review on NIPS 19.
Yunqi Cai
  • VP-flow-lda Finished;
  • 1 try VP-flow without the last layer maping matrix
  • 2 control the class covariance matrix
  • 3 explore the fitting limit of VP-flow
Zhiyuan Tang
  • WaveGlow model for audio augmentation.
  • Test whether the aug helps ASR/VPR.
Lantian Li
  • Fix bugs in fnvpr, and release vpr product packages.
  • Pull request egs of CN-Celeb.
  • Study on flow models.
  • Hire-Flow
Ying Shi
  • experiment about flow denoise
  • finish the experiment and record it on cvss
  • write patent
Wenqiang Du
Haoran Sun
Yue Fan
Jiawen Kang
  • Sent materials to CUHK;
  • Reading paper about model robustness;
  • Kaldi baseline checking.
  • Experiences about CN-Celeb;
  • Preparing for CUHK inverview.
Ruiqi Liu
  • Do experiments on CN-Celeb egs and VGG.
  • tf-kaldi
  • Experiences according to last week's plan
Sitong Cheng
Zhixin Liu
取自“http://index.cslt.org/mediawiki/index.php?title=2019-12-09&oldid=34418”