“2019-02-20”版本间的差异

来自cslt Wiki
跳转至: 导航搜索
第50行: 第50行:
 
* finished x-vector system
 
* finished x-vector system
 
||  
 
||  
* adjust the parameters for situations where the results are bad.
+
* improve the system ,because the results is bad.
 
||
 
||
 
*   
 
*   

2019年2月20日 (三) 04:20的版本

People Last Week This Week Task Tracking (DeadLine)
Yibo Liu
Xiuqi Jiang
Jiayao Wu
  • finished several experiments on node-sparseness
  • keep on doing experiments
Zhaodi Qi
  • finished x-vector system
  • improve the system ,because the results is bad.
Jiawei Yu
Yunqi Cai
  • 1,use the ASR test_data set ref.txt to test the original bert and then compare with the fine-tuned Bert.
  • 2,compare the difference between hyp.text which comes from ASR test results, and investigate how to mask the error in the sentence of the ASR result.
  • 3,set a rule among all the result(hyp.text) to find out the error to be masked.
Dan He
  • Continue to study the time complexity of TT-decomposition
Yang Zhang
Wenwei Dong
  • ivector+fbank as feature compare with fbank in GOP
  • read papers, find some new speaker adaptation methods