“2024-04-22”版本间的差异

来自cslt Wiki
跳转至： 导航、 搜索
 
（11位用户的14个中间修订版本未显示）
第19行： 第19行：
 
|Lantian Li
 
|Lantian Li
 
||
 
||
−
*  
+
* GPU status [https://z1et6d3xtb.feishu.cn/wiki/XGcGwRK5viJmpRkjH9AczIhynCh]
 +
* Projects (AED delivery, TSE plan)
 +
* ASIP-BUPT (NeuralScoring, CohortSS)
 +
* Welcome to Rabbit04
 +
* BlockChain Course
 
||
 
||
 
*  
 
*  
第25行： 第29行：
 
*   
 
*   
 
|-
 
|-
 +
  
  
第30行： 第35行：
 
|Ying Shi
 
|Ying Shi
 
||  
 
||  
−
*
+
* Finish SPL paper
 +
* Discuss about next DI-TING
 +
* Restart Cohort ASR
 +
* [https://z1et6d3xtb.feishu.cn/docx/XdCGdTkarolEFyx9Ig3cCkzBndc?from=from_copylink group work]
 
||
 
||
 
*  
 
*  
第41行： 第49行：
 
|Zhenghai You
 
|Zhenghai You
 
||  
 
||  
−
*  
+
* Paper reading report
 +
* Prepare extreme data for Huawei project( Crawling Chinese karaoke song, then mixed with rock, punk music data as TSE task's noise )
 
||
 
||
 
*  
 
*  
第65行： 第74行：
 
|Chen Chen
 
|Chen Chen
 
||  
 
||  
−
*  
+
* read papers
 +
* vii group [https://z1et6d3xtb.feishu.cn/docx/CNzFdnE0toiDtDxNkq2cetngndf?from=from_copylink]
 +
** Structure have good news
 +
** Strategy need to be checked
 +
** Data collection need more effort
 
||
 
||
 
*  
 
*  
第76行： 第89行：
 
|Xiaolou Li
 
|Xiaolou Li
 
||  
 
||  
−
*  
+
* Experiment on E-Branchformer and Resnet3D + E-Branchformer(last exp on this)
 +
* Paper Reading
 
||
 
||
 
*  
 
*  
第112行： 第126行：
 
|Wan Lin
 
|Wan Lin
 
||  
 
||  
−
*  
+
* Graduation paper revision （already submitted）
 +
* NS
 +
** transfer to wespeaker-toolkit（fix bug）
 +
** all-pairs training & mix training
 
||
 
||
 
*
 
*
第123行： 第140行：
 
|Tianhao Wang
 
|Tianhao Wang
 
||  
 
||  
−
*  
+
* EA-ASP exps [https://z1et6d3xtb.feishu.cn/docx/BywjdkGvNou12sxQ4dAcxYa9noh]
 +
** target and nontarget training data ratio (1:1 to 2:8)
 +
** totally mix training data
 +
* spex+ code modification
 
||
 
||
 
*  
 
*  
第156行： 第176行：
 
|Jiaying Wang
 
|Jiaying Wang
 
||  
 
||  
−
*  
+
* cohort gender-aware verification
 +
** need regenerate training, validation and test data
 +
 
 
||
 
||
 
*  
 
*  
第181行： 第203行：
 
|Wenqiang Du
 
|Wenqiang Du
 
||  
 
||  
−
* EfficientNet-B6 kws model has been trained， Some basic tests have been completed[https://z1et6d3xtb.feishu.cn/wiki/MUsMwvFRli128Mk52NIcKK4qnEg?from=from_copylink]
+
* EfficientNet-B6 kws model has been trained[https://z1et6d3xtb.feishu.cn/wiki/MUsMwvFRli128Mk52NIcKK4qnEg?from=from_copylink]
 
* Large hard negative data collection has been complete 60%
 
* Large hard negative data collection has been complete 60%
 
** （about 5000h collect 5000 FA）
 
** （about 5000h collect 5000 FA）
第194行： 第216行：
 
|Yang Wei
 
|Yang Wei
 
||  
 
||  
−
*  
+
* Evaluate mispronunciation detection system with detection cost function
 +
* Try to estimate best DCF threshold without test set
 
||
 
||
 
*
 
*
第216行： 第239行：
 
||
 
||
 
* Started ASR Data Collection
 
* Started ASR Data Collection
−
** 6.3K collected so far
+
** 6.5K collected so far
 
* Course work
 
* Course work
 
||
 
||
第232行： 第255行：
 
||
 
||
 
*
 
*
 +
||
 +
* 
 +
|-
 +
 +
|-
 +
|Qi Qu
 +
||
 +
* Web service:
 +
** EfficientNetB6-based KWS
 +
* KWS model training:
 +
** FA collected and processed (~40k)
 +
** Appended to training dataset
 +
* Other:
 +
** FunASR w/ hotwords enabled as cloud verification
 +
||
 +
* Test:
 +
** KWS models: B0/B6; alone and combined as two-phased processing
 +
** KWS model + FunASR w/ hotwords
 +
** Different trigger strategies
 +
* KWS model training:
 +
** Collect FA in large scale
 
||
 
||
 
*   
 
*   
 
|-
 
|-

2024年4月22日 (一) 11:11的最后版本

People This Week Next Week Task Tracking (DeadLine)
Dong Wang
  • NSFC XAI proposal
  • AI Graph PPT recheck
  • Course preparation
Lantian Li
  • GPU status [1]
  • Projects (AED delivery, TSE plan)
  • ASIP-BUPT (NeuralScoring, CohortSS)
  • Welcome to Rabbit04
  • BlockChain Course
Ying Shi
  • Finish SPL paper
  • Discuss about next DI-TING
  • Restart Cohort ASR
  • group work
Zhenghai You
  • Paper reading report
  • Prepare extreme data for Huawei project( Crawling Chinese karaoke song, then mixed with rock, punk music data as TSE task's noise )
Junming Yuan
  • Paper reading report prepared
  • FA data analysis[2]
  • AI Graph slides refinement
  • NSFC check
Chen Chen
  • read papers
  • vii group [3]
    • Structure have good news
    • Strategy need to be checked
    • Data collection need more effort
Xiaolou Li
  • Experiment on E-Branchformer and Resnet3D + E-Branchformer(last exp on this)
  • Paper Reading
Zehua Liu
  • papper reading
  • cropsize exp
  • AKVSR code（still doing）
Pengqi Li
  • ICASSP papers reading[4]
  • Start Experiment(PID) on Timit(Extend workshop paper)
  • Leave of Absence(Family matters)
Wan Lin
  • Graduation paper revision （already submitted）
  • NS
    • transfer to wespeaker-toolkit（fix bug）
    • all-pairs training & mix training
Tianhao Wang
  • EA-ASP exps [5]
    • target and nontarget training data ratio (1:1 to 2:8)
    • totally mix training data
  • spex+ code modification
Zhenyu Zhou
  • ICASSP2024 poster
Junhui Chen
  • Graduation paper
Jiaying Wang
  • cohort gender-aware verification
    • need regenerate training, validation and test data
Yu Zhang
  • AutoML
    • learners: [xgboost lgbm xgb_limitdepth rf]
    • use last 30 days metrics to predict the return value
  • RL continuous learning related paper reading
Wenqiang Du
  • EfficientNet-B6 kws model has been trained[6]
  • Large hard negative data collection has been complete 60%
    • （about 5000h collect 5000 FA）
Yang Wei
  • Evaluate mispronunciation detection system with detection cost function
  • Try to estimate best DCF threshold without test set
Lily
  • Overview for thesis
  • AIgraph100 course material
Turi
  • Started ASR Data Collection
    • 6.5K collected so far
  • Course work
Yue Gu
  • Parallel lattice construction
  • Method reproduction, a work in ICASSP2024 related to my contextual ASR
  • Paper reading (50%)
Qi Qu
  • Web service:
    • EfficientNetB6-based KWS
  • KWS model training:
    • FA collected and processed (~40k)
    • Appended to training dataset
  • Other:
    • FunASR w/ hotwords enabled as cloud verification
  • Test:
    • KWS models: B0/B6; alone and combined as two-phased processing
    • KWS model + FunASR w/ hotwords
    • Different trigger strategies
  • KWS model training:
    • Collect FA in large scale
取自“http://index.cslt.org/mediawiki/index.php?title=2024-04-22&oldid=41666”