“2024-08-12”版本间的差异

来自cslt Wiki
跳转至: 导航搜索
 
(9位用户的11个中间修订版本未显示)
第6行: 第6行:
 
|Dong Wang
 
|Dong Wang
 
||
 
||
*
+
* KDD paper done
 +
* ISCSLP review done
 +
* Public talk (primary schools)  done
 +
 
 +
 
 
||
 
||
 
*
 
*
第17行: 第21行:
 
|Lantian Li
 
|Lantian Li
 
||
 
||
*
+
* GPU status [https://z1et6d3xtb.feishu.cn/wiki/XGcGwRK5viJmpRkjH9AczIhynCh]
 +
* Projects
 +
** Proposal of DiTing3.0 Phase 2th
 +
* AI graph
 +
** High school demo lecture
 +
** High school handbook (16/40)
 
||
 
||
*
+
* High school handbook (22/40)
 
||
 
||
*
+
*  
 
|-
 
|-
 +
 +
  
  
第28行: 第39行:
 
|Ying Shi
 
|Ying Shi
 
||
 
||
*  
+
* Text enroll keywords spotting
 +
** cross attention at low layer (95%/4%)
 +
** corss attention at high layer (92%/6%)
 +
* Conditional-Chain overlap ASR
 +
* group work [https://z1et6d3xtb.feishu.cn/docx/Vr5UdIjvvohB6zxr5mXc3X95nJc?from=from_copylink]
 
||
 
||
 
*
 
*
第72行: 第87行:
 
|Xiaolou Li
 
|Xiaolou Li
 
||
 
||
*
+
* Long context test
 +
* Paper reading (memory of LLM)
 
||
 
||
 
*
 
*
第83行: 第99行:
 
|Zehua Liu
 
|Zehua Liu
 
||
 
||
*
+
*CNVSRC 2024 Technical Report and some CNVSRC 2024 stuff
 
||
 
||
 
*
 
*
第105行: 第121行:
 
|Wan Lin
 
|Wan Lin
 
||
 
||
*
+
* prepare training NS model in VoxBlink
 
||
 
||
 
*
 
*
第116行: 第132行:
 
|Tianhao Wang
 
|Tianhao Wang
 
||
 
||
*
+
* prepare VGGSound and WavCaps data
 +
* fix bugs for CLIPSep(ICLR2023) reproducing
 
||
 
||
 
*
 
*
第138行: 第155行:
 
|Junhui Chen
 
|Junhui Chen
 
||
 
||
*
+
* prepare training NS model in VoxBlink
 
||
 
||
 
*
 
*
第149行: 第166行:
 
|Jiaying Wang
 
|Jiaying Wang
 
||
 
||
*
+
* conditional chain on libri2mix(training)
 
||
 
||
 
*
 
*
第160行: 第177行:
 
|Yu Zhang
 
|Yu Zhang
 
||
 
||
*
+
* AED engineering problem assist
 +
** train model with max pooling instead of avg pooling aim to tackle mismatch between training trunk size and test trunk size (no significant improvement)
 +
** data aug with wind noise
 +
* paper reading about (Spatial)TimeSeries-LLM
 
||
 
||
 
*
 
*
第183行: 第203行:
 
|Yang Wei
 
|Yang Wei
 
||
 
||
*
+
* Test KWS model (Chinese+Uyghur+Kazakh; Chinese+Minnan+Uyghur) https://z1et6d3xtb.feishu.cn/docx/MtTcd7xfuoLhcVx3nDocrphinmd?from=create_suite_copy
 
||
 
||
 
*
 
*
第220行: 第240行:
 
|Qi Qu
 
|Qi Qu
 
||
 
||
*  
+
* KWS:
 +
** zh48 (Mandarin Chinese 48 keywords) test dataset annotated.
 +
** Test across models: keyword-wise ROC.
 
||
 
||
 
*
 
*

2024年8月19日 (一) 09:34的最后版本

People This Week Next Week Task Tracking (DeadLine)
Dong Wang
  • KDD paper done
  • ISCSLP review done
  • Public talk (primary schools) done


Lantian Li
  • GPU status [1]
  • Projects
    • Proposal of DiTing3.0 Phase 2th
  • AI graph
    • High school demo lecture
    • High school handbook (16/40)
  • High school handbook (22/40)
Ying Shi
  • Text enroll keywords spotting
    • cross attention at low layer (95%/4%)
    • corss attention at high layer (92%/6%)
  • Conditional-Chain overlap ASR
  • group work [2]
Zhenghai You
Junming Yuan
  • fix some bug in Hubert pretraining experiment
    • clean Hubert pretraining result[3]
Chen Chen
Xiaolou Li
  • Long context test
  • Paper reading (memory of LLM)
Zehua Liu
  • CNVSRC 2024 Technical Report and some CNVSRC 2024 stuff
Pengqi Li
  • Analyze EA-ASP and summarize the recent work.
Wan Lin
  • prepare training NS model in VoxBlink
Tianhao Wang
  • prepare VGGSound and WavCaps data
  • fix bugs for CLIPSep(ICLR2023) reproducing
Zhenyu Zhou
  • Model quantification does not work
Junhui Chen
  • prepare training NS model in VoxBlink
Jiaying Wang
  • conditional chain on libri2mix(training)
Yu Zhang
  • AED engineering problem assist
    • train model with max pooling instead of avg pooling aim to tackle mismatch between training trunk size and test trunk size (no significant improvement)
    • data aug with wind noise
  • paper reading about (Spatial)TimeSeries-LLM
Wenqiang Du
  • Primary school handbook (43/46)
  • Upgrade the model by using real-life FA data
Yang Wei
Lily
Turi
Yue Gu
Qi Qu
  • KWS:
    • zh48 (Mandarin Chinese 48 keywords) test dataset annotated.
    • Test across models: keyword-wise ROC.