“2020-03-23”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33行: 第33行:
 
|Zhiyuan Tang
 
|Zhiyuan Tang
 
||  
 
||  
*
+
* Multi-conditioned glow, on how to involving in spk info.
 
||
 
||
*  
+
* Continue following blow strategies, or adversarial examples.
 
||
 
||
 
*   
 
*   
第44行: 第44行:
 
|Lantian Li
 
|Lantian Li
 
||  
 
||  
*  
+
* Complete the study on back-end scoring (python vs. Kaldi).
 +
* Quantitative analysis on DNF model.
 +
* Obtain the basic result on VAE-based Bayesian scoring.
 
||
 
||
*  
+
* Study on NDA scoring.
 
||
 
||
 
*   
 
*   
第67行: 第69行:
 
|Wenqiang Du
 
|Wenqiang Du
 
||  
 
||  
*
+
*test DNF based ASR  system
 
||
 
||
 
*
 
*
第126行: 第128行:
 
|Sitong Chengs
 
|Sitong Chengs
 
||  
 
||  
*  
+
* Finish training with fbank
 +
* Made some tests on the model
 
||
 
||
*  
+
* Extract more feature with higher dimension
 +
* Make tests on variable models
 
||
 
||
 
*   
 
*   

2020年10月19日 (一) 01:07的最后版本

People This Week Next Week Task Tracking (DeadLine)
Dong Wang
  • More investigation on subspace DNF.
  • Coine the NDA theory, passed some simulation tests.
  • Some ideas on Beyasian scoring and closed-form training for variational Bayes.
  • Drafted the channel-transform paper
  • More investigation on subspace DNF & NDA. They look closely connected.
Yunqi Cai
  • Some experiments on EBMs model
  • Modify the DNF patent
  • Coding on VAE based bayes score model
  • More experiments of VAE based bayes score model
Zhiyuan Tang
  • Multi-conditioned glow, on how to involving in spk info.
  • Continue following blow strategies, or adversarial examples.
Lantian Li
  • Complete the study on back-end scoring (python vs. Kaldi).
  • Quantitative analysis on DNF model.
  • Obtain the basic result on VAE-based Bayesian scoring.
  • Study on NDA scoring.
Ying Shi
  • Summarize result about current model.
  • more analysis about speech enhancement via Flow
  • More experiments based on current result
Wenqiang Du
  • test DNF based ASR system
Haoran Sun
Yue Fan
  • Looking for data sources
  • Upload data and program processing
  • Solve the remaining data collection issues
  • Processing 500 people with the pipeline program
Jiawen Kang
  • Speaker vector distribution experiments.
  • Arranging tf-kaldi codes.
  • Runing meta learning experiments.
Ruiqi Liu
  • Arrang experiments code and observe the distribution of speakers in different scenes.
  • Learn mate-learning and do experiments.
Sitong Chengs
  • Finish training with fbank
  • Made some tests on the model
  • Extract more feature with higher dimension
  • Make tests on variable models
Zhixin Liu
Zhixin Liu
Haolin Chen
  • Some investigation on the structure of glow
  • More experiments
  • Prepare paper