“2025-01-13”版本间的差异

来自cslt Wiki
跳转至： 导航、 搜索
（以“{| class="wikitable" !People !! This Week !! Next Week !! Task Tracking (<font color="red">DeadLine</font>) |- |- |Dong Wang || * || * || * |- |- |Lantian Li ||...”为内容创建页面）
 
 
（17位用户的19个中间修订版本未显示）
第6行： 第6行：
 
|Dong Wang
 
|Dong Wang
 
||
 
||
−
*  
+
* AI handbook high-school version, recheck
 +
* Publication stuff
  
 
||
 
||
第18行： 第19行：
 
|Lantian Li
 
|Lantian Li
 
||
 
||
−
*  
+
* Go on AI-Graph EN (49/50)
 
||
 
||
 
*
 
*
第29行： 第30行：
 
|Ying Shi
 
|Ying Shi
 
||
 
||
−
*  
+
* Try to reproduce the CTC conditional-chain ASR baseline (failed)
 +
* check code with PIT loss (code looks fine)
 +
* Try to train a model that only recognizes the most dominant components (failed)
 
||
 
||
 
*  
 
*  
第41行： 第44行：
 
|Zhenghai You
 
|Zhenghai You
 
||
 
||
−
*  
+
* The first Chinese version of the paper [https://www.overleaf.com/read/hgttnggctwps#2facce]
 +
* Online Demo UI Design for Huawei Project[https://huggingface.co/spaces/swc2/Target-speaker-extraction]
 
||
 
||
 
*
 
*
第51行： 第55行：
 
|Junming Yuan
 
|Junming Yuan
 
||
 
||
−
*  
+
* Check the high school AI handbook(done)
 +
* Organize our photos
 +
* The results of MT-Hubert on LS960[https://z1et6d3xtb.feishu.cn/docx/HUIad3uqgozpEyxIwuTcU9uInsb]
 +
**2-mixed Test: 400K steps top-2 ACC: 74.63% ---> 1600K steps top-2 ACC: 79.43%
 
||
 
||
 
*
 
*
第62行： 第69行：
 
|Xiaolou Li
 
|Xiaolou Li
 
||
 
||
−
*  
+
* Process server code update
 +
* Data audit (until 15th Jan)
 
||
 
||
 
*  
 
*  
第73行： 第81行：
 
|Zehua Liu
 
|Zehua Liu
 
||
 
||
−
*
+
*Paper Reading
 +
*Writing Code for loading larger LLM（32B and 72B）
 +
*Interspeech paper writing
 +
*Collected Data Checking （With Xiaolou）
 
||
 
||
 
*
 
*
第84行： 第95行：
 
|Pengqi Li
 
|Pengqi Li
 
||
 
||
−
*  
+
* IS25 Proposal almost done
 +
** writing the experimental part
 +
** checks and analysis the result
 +
* Go to the hospital for a follow-up and return lab.
 
||
 
||
 
*
 
*
第95行： 第109行：
 
|Wan Lin
 
|Wan Lin
 
||
 
||
−
*  
+
* NS: Adopt multi-enroll, margin-bce and long-duration test to resnet+transformer model: EER 1.23%->1.13%
 
||
 
||
 
*
 
*
第106行： 第120行：
 
|Tianhao Wang
 
|Tianhao Wang
 
||
 
||
−
*  
+
* writing the chain-based sound sep code
 +
* test some SED pretraining under mix scenario
 
||
 
||
 
*
 
*
第117行： 第132行：
 
|Xiaoxue Luo
 
|Xiaoxue Luo
 
||
 
||
−
*  
+
* got a bad flu
 +
* read papers
 
||
 
||
 
*
 
*
第150行： 第166行：
 
|Jiaying Wang
 
|Jiaying Wang
 
||
 
||
−
*
+
* IS2025 proposal[https://z1et6d3xtb.feishu.cn/docx/T8IvdtN32o8kHWxqcPvc5LsWnce]
 
||
 
||
 
*
 
*
第161行： 第177行：
 
|Yu Zhang
 
|Yu Zhang
 
||
 
||
−
*  
+
* Finish Multi Agent Investment pipeline debug, experiment still running (can get a draft result this week)
 
||
 
||
 
*
 
*
第172行： 第188行：
 
|Wenqiang Du
 
|Wenqiang Du
 
||
 
||
−
*  
+
* Check Primary handbook
 +
** Related PPT and Jiaoan
 +
*Some  project cooperation
 
||
 
||
 
*
 
*
第183行： 第201行：
 
|Yang Wei
 
|Yang Wei
 
||
 
||
−
*  
+
* Train Text enroll kws model (pretrain w ctc loss + finetune w/o ctc loss). Not success yet.
 +
* Develop ASR REST service based on FunASR model.
 
||
 
||
 
*
 
*
第193行： 第212行：
 
|Turi
 
|Turi
 
||
 
||
−
*  
+
* Did experiment on Conformer CTC with nbpe=500 (26 to 18 WER)
 +
* Refined ICASSP paper for final submission and submitted it
 +
* Prepare for interview
 
||
 
||
 
*  
 
*  
第201行： 第222行：
 
|Yue Gu
 
|Yue Gu
 
||
 
||
−
*  
+
* design the fine-grained personality extractor to produce the phone-level voice charactor similarity (code is in progress)
 +
* check the primary school handbook
 
||
 
||
 
*
 
*
第210行： 第232行：
 
|Qi Qu
 
|Qi Qu
 
||
 
||
−
*  
+
* AED:
 +
** CED classifiers implemented on mr536 NPU.
 +
* KWS:
 +
** Training data collected and processed for Qingdao dialect, 20 keywords.
 +
** Analysis of some prod FAs.
 +
* Android demo and supporting backend services: KWS + ASR/MT -> instruction submission.
 
||
 
||
 
*
 
*

2025年1月13日 (一) 10:57的最后版本

People This Week Next Week Task Tracking (DeadLine)
Dong Wang
  • AI handbook high-school version, recheck
  • Publication stuff
Lantian Li
  • Go on AI-Graph EN (49/50)
Ying Shi
  • Try to reproduce the CTC conditional-chain ASR baseline (failed)
  • check code with PIT loss (code looks fine)
  • Try to train a model that only recognizes the most dominant components (failed)
Zhenghai You
  • The first Chinese version of the paper [1]
  • Online Demo UI Design for Huawei Project[2]
Junming Yuan
  • Check the high school AI handbook(done)
  • Organize our photos
  • The results of MT-Hubert on LS960[3]
    • 2-mixed Test: 400K steps top-2 ACC: 74.63% ---> 1600K steps top-2 ACC: 79.43%
Xiaolou Li
  • Process server code update
  • Data audit (until 15th Jan)
Zehua Liu
  • Paper Reading
  • Writing Code for loading larger LLM（32B and 72B）
  • Interspeech paper writing
  • Collected Data Checking （With Xiaolou）
Pengqi Li
  • IS25 Proposal almost done
    • writing the experimental part
    • checks and analysis the result
  • Go to the hospital for a follow-up and return lab.
Wan Lin
  • NS: Adopt multi-enroll, margin-bce and long-duration test to resnet+transformer model: EER 1.23%->1.13%
Tianhao Wang
  • writing the chain-based sound sep code
  • test some SED pretraining under mix scenario
Xiaoxue Luo
  • got a bad flu
  • read papers
Zhenyu Zhou
Junhui Chen
Jiaying Wang
  • IS2025 proposal[4]
Yu Zhang
  • Finish Multi Agent Investment pipeline debug, experiment still running (can get a draft result this week)
Wenqiang Du
  • Check Primary handbook
    • Related PPT and Jiaoan
  • Some project cooperation
Yang Wei
  • Train Text enroll kws model (pretrain w ctc loss + finetune w/o ctc loss). Not success yet.
  • Develop ASR REST service based on FunASR model.
Turi
  • Did experiment on Conformer CTC with nbpe=500 (26 to 18 WER)
  • Refined ICASSP paper for final submission and submitted it
  • Prepare for interview
Yue Gu
  • design the fine-grained personality extractor to produce the phone-level voice charactor similarity (code is in progress)
  • check the primary school handbook
Qi Qu
  • AED:
    • CED classifiers implemented on mr536 NPU.
  • KWS:
    • Training data collected and processed for Qingdao dialect, 20 keywords.
    • Analysis of some prod FAs.
  • Android demo and supporting backend services: KWS + ASR/MT -> instruction submission.
取自“http://cslt.org/mediawiki/index.php?title=2025-01-13&oldid=42756”