“ASR Status Report 2017-4-5”版本间的差异

来自cslt Wiki
跳转至: 导航搜索
第37行: 第37行:
 
|Lantian Li   
 
|Lantian Li   
 
||  
 
||  
* Deep speaker : GMM-tnorm and frame-PLDA is effective。
+
* Deep speaker : GMM-tnorm and frame-PLDA is effective.
 
* LM adaptation for Accent ASR.  
 
* LM adaptation for Accent ASR.  
 
||  
 
||  

2017年4月5日 (三) 08:25的版本


Date People Last Week This Week
2017.4.5


Yanqing Wang
  • read source code of thchs30 and run the shell scripts
  • continue on the DNN-HMM and source code of thchs30
Ying Shi
  • baseline of swbd speech recognition and partial collaborative training of new joint structure(enhance speech task) finished
  • baseline of fisher SRE partial collaborative training of new joint structure(enhance speaker task)
Yixiang Chen
Lantian Li
  • Deep speaker : GMM-tnorm and frame-PLDA is effective.
  • LM adaptation for Accent ASR.
  • Deep speaker : phone-aware
Zhiyuan Tang
  • WSJ AM assists AP-OLR lid. results
  • noise robustness analysis, etc.




Date People Last Week This Week
2017.3.27


Jingyi Lin
Yanqing Wang
Hang Luo
Ying Shi
  • baseline of swbd speech recognition and partial collaborative training of new joint structure
  • go on this task
Yixiang Chen
Lantian Li
  • Deep speaker : GMM-UBM, frame-PLDA
  • Check Springer book.
  • LM adaptation for Accent ASR.
  • Deep speaker : phone-aware
Zhiyuan Tang
  • LID on more languages, bigger acoustic models
  • continue the experiments