“2025-03-17”版本间的差异

来自cslt Wiki
跳转至： 导航、 搜索
 
（10位用户的13个中间修订版本未显示）
第18行： 第18行：
 
|Lantian Li
 
|Lantian Li
 
||
 
||
−
*  
+
* Submit of the AI-Graph EN version
 +
* Review Master theses
 
||
 
||
 
*
 
*
第66行： 第67行：
 
|Xiaolou Li
 
|Xiaolou Li
 
||
 
||
−
*  
+
* Writing NSFC document
 +
* Find solutions for the low training speed with hybrid data disk
 
||
 
||
 
*  
 
*  
第77行： 第79行：
 
|Zehua Liu
 
|Zehua Liu
 
||
 
||
−
*
+
*Writing NFSC document
 +
*Reading VTS paper
 
||
 
||
 
*
 
*
第88行： 第91行：
 
|Pengqi Li
 
|Pengqi Li
 
||
 
||
−
*  
+
* Prepare the AI course for Tsinghua University middle School.
 +
* Add references to the middle handbook(Done)
 +
* Check middle handbook(1/3)
 
||
 
||
 
*
 
*
第99行： 第104行：
 
|Wan Lin
 
|Wan Lin
 
||
 
||
−
*  
+
* train ns(multi-enroll) for voxblink1+voxceleb2: perform better than before, but still worse than vc2-only
 +
* retrain multi-scenario ns(multi-enroll) model: consistently better than before now [https://z1et6d3xtb.feishu.cn/docx/MxBNdPbLao0tsoxkBVCcUgUoneh?from=from_copylink]
 +
* run ablation experiments: w/o multi-enroll/multi-head/two encoder
 
||
 
||
 
*
 
*
第110行： 第117行：
 
|Tianhao Wang
 
|Tianhao Wang
 
||
 
||
−
*  
+
* sound sep subset training & testing
 +
* huawei interview
 +
* professor Guo's project investigation
 
||
 
||
 
*
 
*
第121行： 第130行：
 
|Xiaoxue Luo
 
|Xiaoxue Luo
 
||
 
||
−
*  
+
* Sound separation
 +
** train AudioSep on subset data and evaluate it with our testing dataset
 +
** read a paper about USS, try to test this model using our test set firstly(adjusting code in progress)
 
||
 
||
 
*
 
*
第132行： 第143行：
 
|Zhenyu Zhou
 
|Zhenyu Zhou
 
||
 
||
−
*
+
* Revise graduation thesis
 
||
 
||
 
*
 
*
第155行： 第166行：
 
|Jiaying Wang
 
|Jiaying Wang
 
||
 
||
−
*
+
* re-design the network structure[https://z1et6d3xtb.feishu.cn/docx/TUHldiaoQoYBqux7JEhcaCXenzh]
 +
** network without cross attention is already quite large and requires more than 2 GPUs for training,need to be reduced
 
||
 
||
 
*
 
*
第166行： 第178行：
 
|Yu Zhang
 
|Yu Zhang
 
||
 
||
−
*  
+
* Some software structure design and function time flow design in Royal Flush (for external robot project)
 +
* AED stream inference and score smooth window size experiment
 
||
 
||
 
*
 
*
第190行： 第203行：
 
|Yang Wei
 
|Yang Wei
 
||
 
||
−
*  
+
* Text enroll kws adaptation experiment on cross lingual data. (not work. trying it with more complex adaptation layer)
 
||
 
||
 
*
 
*
第201行： 第214行：
 
||
 
||
 
* Prepared ppt and poster for ICASSP2025
 
* Prepared ppt and poster for ICASSP2025
−
* Presentation video
+
* Did presentation video and submitted with poster
 
||
 
||
 
*  
 
*  

2025年3月17日 (一) 10:58的最后版本

People This Week Next Week Task Tracking (DeadLine)
Dong Wang
  • AI handbook refinment for the college version.
Lantian Li
  • Submit of the AI-Graph EN version
  • Review Master theses
Ying Shi
  • thesis
  • some school-related stuff
Zhenghai You
  • Complete the training and experimentation of the TSE-IRA model[1] for Huawei, and will write the report
Junming Yuan
  • prepare report
  • double-check middle-school AI handbook(1/7)
  • write paper
  • double-check middle-school AI handbook
  • AI practice handbook design of primary school and middle school
Xiaolou Li
  • Writing NSFC document
  • Find solutions for the low training speed with hybrid data disk
Zehua Liu
  • Writing NFSC document
  • Reading VTS paper
Pengqi Li
  • Prepare the AI course for Tsinghua University middle School.
  • Add references to the middle handbook(Done)
  • Check middle handbook(1/3)
Wan Lin
  • train ns(multi-enroll) for voxblink1+voxceleb2: perform better than before, but still worse than vc2-only
  • retrain multi-scenario ns(multi-enroll) model: consistently better than before now [2]
  • run ablation experiments: w/o multi-enroll/multi-head/two encoder
Tianhao Wang
  • sound sep subset training & testing
  • huawei interview
  • professor Guo's project investigation
Xiaoxue Luo
  • Sound separation
    • train AudioSep on subset data and evaluate it with our testing dataset
    • read a paper about USS, try to test this model using our test set firstly(adjusting code in progress)
Zhenyu Zhou
  • Revise graduation thesis
Junhui Chen
  • Finished diarization test in vox-o
    • continue in vox-e
Jiaying Wang
  • re-design the network structure[3]
    • network without cross attention is already quite large and requires more than 2 GPUs for training,need to be reduced
Yu Zhang
  • Some software structure design and function time flow design in Royal Flush (for external robot project)
  • AED stream inference and score smooth window size experiment
Wenqiang Du
  • Check Primary handbook V3.1
  • Check middle handbook
  • We have Dragon05 (A6000 * 6)
Yang Wei
  • Text enroll kws adaptation experiment on cross lingual data. (not work. trying it with more complex adaptation layer)
Turi
  • Prepared ppt and poster for ICASSP2025
  • Did presentation video and submitted with poster
Yue Gu
  • synthesize 100h for each target spk and use KL loss as the regular term, the CER of target speakers reduce 10%.
Qi Qu
  • Android demo of Text-enroll KWS model.
取自“http://index.cslt.org/mediawiki/index.php?title=2025-03-17&oldid=42928”