“2024-12-02”版本间的差异

来自cslt Wiki
跳转至： 导航、 搜索
 
（7位用户的7个中间修订版本未显示）
第6行： 第6行：
 
|Dong Wang
 
|Dong Wang
 
||
 
||
−
*
+
* Public talk in Chen Zhou
 +
* 2nd teacher training in Tsinghua middle school
 
||
 
||
 
*
 
*
第17行： 第18行：
 
|Lantian Li
 
|Lantian Li
 
||
 
||
−
*
+
* Check AI handbook high school (4/40)
 +
* Go on AI-Graph EN Chapter 4 (37/50)
 +
* Polish 2025 Daily Sign (5/12)
 +
* ICASSP Rebuttal
 
||
 
||
 
*
 
*
第63行： 第67行：
 
|Xiaolou Li
 
|Xiaolou Li
 
||
 
||
−
*
+
* Data Condition
 +
** CVS3: 2/4
 +
** GA data: ready for training
 +
* VSR-LLM Align with contrastive loss:
 +
** Loss drop great, but the output of Connector still strange, need to modify the function of align
 +
* thesis proposal
 +
* paper organization about VSR SSL method (for proposal)
 
||
 
||
 
*
 
*
第74行： 第84行：
 
|Zehua Liu
 
|Zehua Liu
 
||
 
||
−
*
+
*Label corruption still need explore
 +
*Inference Hyper-Parameter change
 +
**Iter-1（45.52%）-> Iter-2（44.60%）-> Iter-3 (44.30%)
 +
*Prepare start-term report and the draft of graduation thesis
 
||
 
||
 
*
 
*
第96行： 第109行：
 
|Wan Lin
 
|Wan Lin
 
||
 
||
−
*
+
* NS: all transformer
 +
** 5w spk + multi-enroll: EER 1.67%
 +
** variable-length training & adding spk
 
||
 
||
 
*
 
*
第130行： 第145行：
 
|Zhenyu Zhou
 
|Zhenyu Zhou
 
||
 
||
−
*
+
*Prepare mid-term report and the draft of graduation thesis
 
||
 
||
 
*
 
*
第178行： 第193行：
 
||
 
||
 
* Project cooperation communication
 
* Project cooperation communication
−
** Tencent  
+
** Tencent-Contract discuss
 
** dongxin- technical report about vpr
 
** dongxin- technical report about vpr
 
** zhongchuan - local asr
 
** zhongchuan - local asr
第193行： 第208行：
 
|Yang Wei
 
|Yang Wei
 
||
 
||
−
*
+
* Train text enroll kws model with 7000h data with spec aug[https://z1et6d3xtb.feishu.cn/docx/WLLLd2zQvoiE78xfMhpcHejdnnd]
 +
* Experiment about inference rtf.[https://z1et6d3xtb.feishu.cn/docx/Sgo3dCvmqoovXOxvneYcBnXOnoh]
 
||
 
||
 
*
 
*
第213行： 第229行：
 
|Turi
 
|Turi
 
||
 
||
−
*
+
* Paper reading
 
||
 
||
 
*
 
*

2024年12月2日 (一) 10:58的最后版本

People This Week Next Week Task Tracking (DeadLine)
Dong Wang
  • Public talk in Chen Zhou
  • 2nd teacher training in Tsinghua middle school
Lantian Li
  • Check AI handbook high school (4/40)
  • Go on AI-Graph EN Chapter 4 (37/50)
  • Polish 2025 Daily Sign (5/12)
  • ICASSP Rebuttal
Ying Shi
  • Some personal stuff
  • Find some numerical issues of current conditional model (will be fixed in this week)
Zhenghai You
  • Huawei TSE IRA Demo[1]
Junming Yuan
  • Some varient experiments with MT-HuBERT[2]
  • get familiar with text-enroll experiment
Xiaolou Li
  • Data Condition
    • CVS3: 2/4
    • GA data: ready for training
  • VSR-LLM Align with contrastive loss:
    • Loss drop great, but the output of Connector still strange, need to modify the function of align
  • thesis proposal
  • paper organization about VSR SSL method (for proposal)
Zehua Liu
  • Label corruption still need explore
  • Inference Hyper-Parameter change
    • Iter-1（45.52%）-> Iter-2（44.60%）-> Iter-3 (44.30%)
  • Prepare start-term report and the draft of graduation thesis
Pengqi Li
Wan Lin
  • NS: all transformer
    • 5w spk + multi-enroll: EER 1.67%
    • variable-length training & adding spk
Tianhao Wang
  • Some test for out-of-domain text query gap [3]
  • thesis proposal
  • thesis proposal
Xiaoxue Luo
  • production of 2025 Daily Sign(5/12)
  • production of 2025 Daily Sign(12/12)
Zhenyu Zhou
  • Prepare mid-term report and the draft of graduation thesis
Junhui Chen
Jiaying Wang
Yu Zhang
  • thesis proposal
  • paper reading about Multi-Agent Collaboration
    • Cooperation in Different Roles
    • Competition in the Same Role
Wenqiang Du
  • Project cooperation communication
    • Tencent-Contract discuss
    • dongxin- technical report about vpr
    • zhongchuan - local asr
    • Heeryuan - In bidding about RAG
Yang Wei
  • Train text enroll kws model with 7000h data with spec aug[4]
  • Experiment about inference rtf.[5]
Lily
Turi
  • Paper reading
Yue Gu
  • ablation study about adaptation data, badcase analysis [6]
  • read several papers about synthetic data for data argumentation
Qi Qu
  • Finding ideal thresholds for `zh48+henan22` and `zh48+chongqing15` KWS models. `zh48+chongqing15` performs poorly and needs re-arranging word list and retraining.
  • NPU development/test environment setup: mr536 and v851se (new).
取自“http://cslt.org/mediawiki/index.php?title=2024-12-02&oldid=42591”