“2024-04-15”版本间的差异

来自cslt Wiki
跳转至: 导航搜索
 
(19位用户的35个中间修订版本未显示)
第6行: 第6行:
 
|Dong Wang
 
|Dong Wang
 
||  
 
||  
*  
+
* Interspeech review
 +
* AI primary education design
 +
* <Illustraitver AI> slides refinement
 +
 
 
||
 
||
 
*
 
*
第17行: 第20行:
 
|Lantian Li
 
|Lantian Li
 
||
 
||
*  
+
* GPU status [https://z1et6d3xtb.feishu.cn/wiki/XGcGwRK5viJmpRkjH9AczIhynCh]
 +
* Projects (POC of Cough/Humming detection, TSE proposal)
 +
* ASIP-BUPT (NeuralScoring, CohortTSE)
 +
* AI Course Polish
 +
* BlockChain Course
 
||
 
||
 
*  
 
*  
第28行: 第35行:
 
|Ying Shi
 
|Ying Shi
 
||  
 
||  
*
+
* Finish SPL paper version1
 +
* Wake Up model test with real speech
 +
* Chinese Phrase-Guided ASR model test with real speech [https://z1et6d3xtb.feishu.cn/wiki/YU81wRCdTijmUxkQWtTcApJTnXf?from=from_copylink group work]
 
||
 
||
 
*  
 
*  
第39行: 第48行:
 
|Zhenghai You
 
|Zhenghai You
 
||  
 
||  
*  
+
* Retrain a SpEx+ model more suitable for online
 +
* Reflect on cohort and reorganized document
 +
* paper reading
 
||
 
||
 
*  
 
*  
第49行: 第60行:
 
|Junming Yuan
 
|Junming Yuan
 
||  
 
||  
*
+
* prepare materials of live broadcast
 +
* paper reading
 +
* got sick
 
||
 
||
*
+
*  
 
||
 
||
 
*   
 
*   
第71行: 第84行:
 
|Xiaolou Li
 
|Xiaolou Li
 
||  
 
||  
*  
+
* Experiment [https://z1et6d3xtb.feishu.cn/docx/Q3C6dcms2ouhUjxZbkycIZ3Jnad?from=from_copylink]
 +
** Resnet3D + Branchformer
 +
** E-Branchformer V2
 +
** Mamba
 +
* Paper Reading
 +
** VSR in ICASSP2024
 +
** Mamba and related paper
 
||
 
||
*  
+
* Continue to fix the bug of e branchformer
 +
* fix the bug of mamba encoder
 +
* paper reading
 
||
 
||
 
*   
 
*   
第82行: 第103行:
 
|Zehua Liu
 
|Zehua Liu
 
||  
 
||  
*
+
*auxiliary loss exp[https://z1et6d3xtb.feishu.cn/docx/Q3C6dcms2ouhUjxZbkycIZ3Jnad?from=from_copylink]
 +
*crop_size exp(still training)
 +
*read papper
 
||
 
||
*  
+
* reproduce other architecture
 
||
 
||
 
*   
 
*   
第93行: 第116行:
 
|Pengqi Li
 
|Pengqi Li
 
||   
 
||   
*  
+
* speech XAI review v1[https://z1et6d3xtb.feishu.cn/docx/E1ZsdHfgJoOfp7xmpTsccGjhnuf]
 +
* polished poster
 +
* live broadcast
 
||
 
||
*
+
* read ICASSP papers
 
||
 
||
 
*   
 
*   
第110行: 第135行:
 
** effect of time of training sample
 
** effect of time of training sample
 
** inherit hard speaker sample
 
** inherit hard speaker sample
* SOTA exploration
 
 
** add channel-time attention in ResNet to get enroll-aware test feature
 
** add channel-time attention in ResNet to get enroll-aware test feature
 
||
 
||
第120行: 第144行:
 
|Tianhao Wang
 
|Tianhao Wang
 
||  
 
||  
*  
+
* EA-ASP exps
 +
** aligned toolkit (wespeaker To sunine), failed. we will align to wespeaker
 +
** aligned training data (weak overlap To strong overlap)
 +
*** concat and weak_overlap worse, overlap and mix better compared to previous
 +
*** NS arch. has advantages under mix, but non under other tests compared to EA-ASP
 
||
 
||
*  
+
* read paper
 +
* reproduce SpEx+ in TSV
 
||
 
||
 
*   
 
*   
第142行: 第171行:
 
|Junhui Chen
 
|Junhui Chen
 
||
 
||
*  
+
* Neural Scoring result [https://z1et6d3xtb.feishu.cn/docx/XUDOdDSg4o0PRZxHfOWcRnKrnJb?from=from_copylink]
 
||
 
||
*
+
* Graduation paper
 
||
 
||
 
*   
 
*   
第153行: 第182行:
 
|Jiaying Wang
 
|Jiaying Wang
 
||  
 
||  
*  
+
* re-organized document 0025[https://z1et6d3xtb.feishu.cn/drive/folder/NsvKfTKmMl4TGBdfKCNcmdFOnie]
 +
* SS_based: Conv-Tasnet with one fixed cohort(still training)[https://z1et6d3xtb.feishu.cn/docx/IeIydyjzJozfUWxvmNfcWASBngg]
 +
* paper reading
 
||
 
||
 
*  
 
*  
第164行: 第195行:
 
|Yu Zhang
 
|Yu Zhang
 
||
 
||
*  
+
* SAC model training and backtesting [https://z1et6d3xtb.feishu.cn/docx/LqT5dwx1boABqLxNYJycj5K5nQg]
 +
* financial quantile work pipeline finish
 
||
 
||
*
+
* add more data to pipeline (more benchmark and more training testing data range)
 +
* append financial-pipeline design and implement detail doc
 +
* AutoML for stock return regression
 
||
 
||
 
*   
 
*   
第175行: 第209行:
 
|Wenqiang Du
 
|Wenqiang Du
 
||  
 
||  
*  
+
* Efficient-B6 pretrain model training
 +
* hard negative training
 +
** FA data is being collected
 
||
 
||
 
*
 
*
第186行: 第222行:
 
|Yang Wei
 
|Yang Wei
 
||  
 
||  
*  
+
* Children mispronunciation detection
 +
** Analyze and check the baseline model
 +
* Huilan
 +
** ASR service bug fix and update
 
||
 
||
 
*
 
*
第195行: 第234行:
 
|-
 
|-
 
|Lily
 
|Lily
 +
||
 +
* Data Analysis / Annotation(金富,刘欢)
 +
* Paper reading [https://z1et6d3xtb.feishu.cn/sheets/WJjEspdCShnFRmt2r92cH3ubnBg?from=from_copylink]
 +
* Assisted to prepare AI graph course materials
 +
* Prepare for AI radiance live broadcast
 +
||
 +
* review v1[https://z1et6d3xtb.feishu.cn/docx/L0jGdCqEXouL8hx8kelcrJzjn8d?from=from_copylink]
 +
||
 +
 +
|-
 +
 +
|-
 +
|Turi
 +
||
 +
* Data collection App[https://uestc.feishu.cn/docx/VvoJdvnZQozvdpxuUB9c8hdun5j?from=from_copylink]
 +
** Tested the app and Fixed some bugs
 +
||
 +
* Start Recording
 +
||
 +
 +
|-
 +
 +
|-
 +
|Qi Qu
 +
||
 +
* Server-side KWS:
 +
** Tested with EfficientNetB{2,4,6,8}
 +
** Service to be implemented with EfficientNetB6 (preview)
 
||
 
||
 
*  
 
*  
 
||
 
||
*
+
 +
|-
 +
 
 +
|-
 +
|Yue Gu
 +
||
 +
* The most experiments of contextual ASR
 +
* The pseudocodes of group stage and bias-phrase decoding lattice
 +
* Read one paper
 +
||
 +
*  
 
||
 
||
 
*   
 
*   
 
|-
 
|-

2024年4月15日 (一) 11:26的最后版本

People This Week Next Week Task Tracking (DeadLine)
Dong Wang
  • Interspeech review
  • AI primary education design
  • <Illustraitver AI> slides refinement
Lantian Li
  • GPU status [1]
  • Projects (POC of Cough/Humming detection, TSE proposal)
  • ASIP-BUPT (NeuralScoring, CohortTSE)
  • AI Course Polish
  • BlockChain Course
Ying Shi
  • Finish SPL paper version1
  • Wake Up model test with real speech
  • Chinese Phrase-Guided ASR model test with real speech group work
Zhenghai You
  • Retrain a SpEx+ model more suitable for online
  • Reflect on cohort and reorganized document
  • paper reading
Junming Yuan
  • prepare materials of live broadcast
  • paper reading
  • got sick
Chen Chen
Xiaolou Li
  • Experiment [2]
    • Resnet3D + Branchformer
    • E-Branchformer V2
    • Mamba
  • Paper Reading
    • VSR in ICASSP2024
    • Mamba and related paper
  • Continue to fix the bug of e branchformer
  • fix the bug of mamba encoder
  • paper reading
Zehua Liu
  • auxiliary loss exp[3]
  • crop_size exp(still training)
  • read papper
  • reproduce other architecture
Pengqi Li
  • speech XAI review v1[4]
  • polished poster
  • live broadcast
  • read ICASSP papers
Wan Lin
  • Finish graduation paper
  • Explore multi-speaker training in NS (how to get batter result in all condition)
    • use wespeaker toolkit
    • effect of time of training sample
    • inherit hard speaker sample
    • add channel-time attention in ResNet to get enroll-aware test feature
Tianhao Wang
  • EA-ASP exps
    • aligned toolkit (wespeaker To sunine), failed. we will align to wespeaker
    • aligned training data (weak overlap To strong overlap)
      • concat and weak_overlap worse, overlap and mix better compared to previous
      • NS arch. has advantages under mix, but non under other tests compared to EA-ASP
  • read paper
  • reproduce SpEx+ in TSV
Zhenyu Zhou
Junhui Chen
  • Neural Scoring result [5]
  • Graduation paper
Jiaying Wang
  • re-organized document 0025[6]
  • SS_based: Conv-Tasnet with one fixed cohort(still training)[7]
  • paper reading
Yu Zhang
  • SAC model training and backtesting [8]
  • financial quantile work pipeline finish
  • add more data to pipeline (more benchmark and more training testing data range)
  • append financial-pipeline design and implement detail doc
  • AutoML for stock return regression
Wenqiang Du
  • Efficient-B6 pretrain model training
  • hard negative training
    • FA data is being collected
Yang Wei
  • Children mispronunciation detection
    • Analyze and check the baseline model
  • Huilan
    • ASR service bug fix and update
Lily
  • Data Analysis / Annotation(金富,刘欢)
  • Paper reading [9]
  • Assisted to prepare AI graph course materials
  • Prepare for AI radiance live broadcast
Turi
  • Data collection App[11]
    • Tested the app and Fixed some bugs
  • Start Recording
Qi Qu
  • Server-side KWS:
    • Tested with EfficientNetB{2,4,6,8}
    • Service to be implemented with EfficientNetB6 (preview)
Yue Gu
  • The most experiments of contextual ASR
  • The pseudocodes of group stage and bias-phrase decoding lattice
  • Read one paper