“ASR Status Report 2017-1-16”版本间的差异

来自cslt Wiki
跳转至: 导航搜索
 
(2位用户的4个中间修订版本未显示)
第42行: 第42行:
 
* find suffix of kazakh
 
* find suffix of kazakh
 
* new language model(6-gram) ppl on original test is 116 and in newtest is 14
 
* new language model(6-gram) ppl on original test is 116 and in newtest is 14
 +
* wer after split suffix is 15.16% without which is 14.43%
 
||  
 
||  
 
* finish the kazak speech recongnition job
 
* finish the kazak speech recongnition job
第60行: 第61行:
 
|Lantian Li   
 
|Lantian Li   
 
||  
 
||  
* Deep speaker embedding.
+
* Deep speaker embedding.[http://192.168.0.51:5555/cgi-bin/cvss/cvss_request.pl?account=lilt&step=view_request&cvssid=591 591]
* Replay detection.
+
* Replay detection.[http://192.168.0.51:5555/cgi-bin/cvss/cvss_request.pl?account=lilt&step=view_request&cvssid=590 590]
 
* Write book.
 
* Write book.
 
||  
 
||  
第75行: 第76行:
 
||  
 
||  
 
* data preparation (phone list, lexicon) for Cantonese ASR
 
* data preparation (phone list, lexicon) for Cantonese ASR
 +
* applying for the visa
 
||  
 
||  
* babel baseline.
+
* babel baseline
 
|-
 
|-
  

2017年1月16日 (一) 05:48的最后版本


Date People Last Week This Week
2017.1.16


Jingyi Lin
Yanqing Wang
Hang Luo
Ying Shi
  • find suffix of kazakh
  • new language model(6-gram) ppl on original test is 116 and in newtest is 14
  • wer after split suffix is 15.16% without which is 14.43%
  • finish the kazak speech recongnition job
Yixiang Chen
Lantian Li
  • Deep speaker embedding.591
  • Replay detection.590
  • Write book.
  • Deep speaker embedding.
  • Replay detection.
  • Write book.
Zhiyuan Tang
  • data preparation (phone list, lexicon) for Cantonese ASR
  • applying for the visa
  • babel baseline


Date People Last Week This Week
2017.1.10


Jingyi Lin
Yanqing Wang
Hang Luo
  • TRP and book checked.
  • Train Japanese ASR baseline
Ying Shi
  • kazak speech recognition version1.0
  • lexicon about Arabic numerals and loan words
  • try to make language model by roots and affixes
Yixiang Chen
  • Integrate CNN + max-margin.
  • test on small data set
  • test CNN + max-margin
  • optimize code
Lantian Li
  • Deep speaker embedding
  • Write book.
  • Integrate TRPs.
  • Deep speaker embedding.
  • Write book.
Zhiyuan Tang
  • checked and submitted several TRPs (including interns');
  • restart joint training;
  • weekly reading.