“2024-07-01”版本间的差异

来自cslt Wiki
跳转至: 导航搜索
第83行: 第83行:
 
|Chen Chen
 
|Chen Chen
 
||
 
||
*
+
* graduate
 +
* entropy (with avhubert/hubert for LRS3/GRID)
 
||
 
||
*
+
* finish ISCSLP paper
 
||
 
||
 
*
 
*

2024年7月1日 (一) 10:58的版本

People This Week Next Week Task Tracking (DeadLine)
Dong Wang
  • A trial talk in Retired Professor Association of THU.
  • Check Aigraph slides
Lantian Li
  • GPU status [1]
  • Projects
    • AED -> model miniaturization/streaming
    • TSE -> 1st phase delivery (this week)
    • VSR -> no progress
    • Finance -> no progress
  • Papers
    • NeuralScoring
  • AI graph
    • Slides checking (10/50)
    • High school handbook (2/40)
Ying Shi
  • Text enroll Keyword Spotting
    • Training with sufficient data augmentation [failed]
  • Cohort Conditional Chain Overlap ASR
    • Fix the bug about the positional embedding of Condition, the performance is still not good
  • Review several NC paper
  • Text enroll Keyword Spotting
    • Employ Homo loss
  • Cohort Conditional Chain Overlap ASR
    • Check the SID performance of the current speaker embedding model on the training dataset
    • Reproduce the previous Cohort-SOT methods on the training dataset
  • Finish the SPL response
Zhenghai You
  • Complete the project deliverables
Junming Yuan
  • find the bug in SSL model finetuning experiment with multi-lingual
    • double check result in [2]
    • The results need check again.
Chen Chen
  • graduate
  • entropy (with avhubert/hubert for LRS3/GRID)
  • finish ISCSLP paper
Xiaolou Li
  • Some exps on Bimamba (parameters search) [3]
  • Paper reading
    • MLLM survey, pre-train, fine-tune and modality alignment
Zehua Liu
  • VSP-LLM Reproduce(LRS3(30h) wer:36.32 > wer: 29.2)[4]
  • still need work on the code
Pengqi Li
Wan Lin
  • Neural Scoring
    • trials test: vox1-e, vox1-h
    • cn: ns(all-genres and 3-genres fine-tuning)
    • variable-chunk training(2-10s):
      training, looks similar to the results of 4/6s
    • weekly report
Tianhao Wang
  • Neural Scoring [5]:
    • vox: vox1-e, vox1-h test [6]
    • cn: three genres fine-tuning: resnet and ns, 2s and 4s
    • weekly report
Zhenyu Zhou
  • Huawei Project Submission
Junhui Chen
  • Neural Scoring [7]:
    • vox: vox1-e, vox1-h test [8]
    • vox: one transformer encoder layer training
    • cn: three genres fine-tuning: resnet and ns, 2s and 4s
Jiaying Wang
  • Preliminary validation: cohort works[9]
Yu Zhang
  • Finance
    • Data Collection (2015 - 2019 HS300 stocks)
  • AED
    • 8k 2s CNN model training and Window inference code
Wenqiang Du
  • Quantified the kws model of Aibabel
  • Training dialect models for AIbabel( Uyghur language)
  • Train a joint model for Chinese, Uyghur, and Kazakh languages
Yang Wei
  • AIbabel
    • Learn to train and test KWS model
Lily
  • Assisted with design of AI courses for primary, high school
  • Some chores about 'AIradiance'
    • Prepare application for Beijing Science and Technology Progress Award
    • Live broadcast
    • Prepare the content for the daily sign poster for July, August
Turi
  • Thesis Proposal Defense
  • Data Collection
    • 31K utterences so far
Yue Gu
  • Prepare the live content
  • writing paper
Qi Qu
  • AED:
    • CED + Linear: c/jni/python lib development and test.
  • AED:
    • CED: Linear to be trained on data.
    • On-device demo.