“Asr-nsfc-weekly-2017-09-25”版本间的差异

来自cslt Wiki
跳转至: 导航搜索
第16行: 第16行:
 
||  
 
||  
 
* correct Kazak speech corpora problems.
 
* correct Kazak speech corpora problems.
* improve Kirghiz text quality, especially spelling mistakes .
+
* improve Kirghiz text corpus quality, especially spelling mistakes .
 
* work on multilingual morpheme segmenter tool for three languages.
 
* work on multilingual morpheme segmenter tool for three languages.
 
|-
 
|-

2017年9月25日 (一) 05:46的版本

Date People Last Week This Week
2017.09.18
清华
  • apply fake Lid into Gsoftmax model(in progress)
  • finish fake Lid decoding
新大
  • acoustic normalization tool, and parallel phonemes for 3 languages is prepared, and used for spell checking and acoustic dictionary building.
  • correct Kazak speech corpora problems.
  • improve Kirghiz text corpus quality, especially spelling mistakes .
  • work on multilingual morpheme segmenter tool for three languages.
民大

Date People Last Week This Week
2017.09.18
清华
  • group-based softmax 模型训练完毕效果得到了相应的提升here
  • Lid + group based softmax
新大
民大
  • 手机录音程序实现基本功能,需进一步完善
  • 蒙语词典校对(正在进行)
  • 完善手机录音程序
  • 拉萨藏语字典部分国标音标的校准
  • 蒙语词典校对