2024-08-05

来自cslt Wiki
2024年8月5日 (一) 10:58Lixl讨论 | 贡献的版本

(差异) ←上一版本 | 最后版本 (差异) | 下一版本→ (差异)
跳转至: 导航搜索
People This Week Next Week Task Tracking (DeadLine)
Dong Wang
  • AC-SQL paper for KDD
  • Several public talks
  • Review for ISCSLP
Lantian Li
  • GPU status [1]
  • AI graph
    • QA slides checked
  • High school handbook (20/40)
Ying Shi
Zhenghai You
  • Fix mutil-scale loss bug in Ex-former(u-net)
  • Tse Project: The performance of the pre-trained model on 12 spk data is poor
  • Writing ICCIP2024 & Complete the experiment
Junming Yuan
  • Hubert pretraining exp(still have problem)
    • pretrain on our libri-keyword dataset(~277h) and finetune on 15-shot GSC dataset with MT
      • top-1 acc --> 9.72%, EER --> 49.73%
    • pretrained model still have problem(Maybe audio and pseudo-label duration differ too much)
Xiaolou Li
  • Reproduce Grouping ViT as the modality projection (trouble in inference)
  • Some test on different prompt from ASR paper. [2]
  • Paper reading (mainly about ASR + LLM and multimodality projection method)
Zehua Liu
  • CNVSRC 2024 things
  • Reading some Speech-separation papper
Pengqi Li
  • Analysis ongoing for pooling with condition(difficult to explain)
Wan Lin
  • NS paper: Supplement experimental results and citations
Tianhao Wang
  • reproducing sound filter (data and code)
  • project things
Zhenyu Zhou
  • Model quantification
Junhui Chen
  • Neural Scoring
    • Revising paper
    • Supplement experiments(finished)
Jiaying Wang
  • reproducing Condition chain code
Yu Zhang
Wenqiang Du
  • primary school handbook (35/46)
Yang Wei
  • Test KWS model with test set v1.0 (result analysis in progress)
Lily
  • Prepare for high shcool summer trip class(last Sunday)
  • Prepare for teacher's course (On this Saturday)
  • AIradiance's daily work
Turi
Yue Gu
  • writing paper
  • read several accent adaptation papers
Qi Qu
  • AED:
    • Classifier trained on "cries" samples.
    • Artificial recall test datasets for "slaps" and "cries".
  • KWS:
    • Mandarin Chinese 48-word recall test dataset: 10 speakers * 10 repeats expected.
  • Misc:
    • Live talk preparation.