“2025-01-20”版本间的差异

来自cslt Wiki
跳转至： 导航、 搜索
 
（13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 Graph high-school version, still on going
  
 
||
 
||
第29行： 第29行：
 
|Ying Shi
 
|Ying Shi
 
||
 
||
−
*  
+
* Huawei TSE online demo design with Zhenghai
 +
* Continue work on conditional overlap ASR model
 
||
 
||
 
*  
 
*  
第41行： 第42行：
 
|Zhenghai You
 
|Zhenghai You
 
||
 
||
−
*  
+
* TSE online service for Huawei project demonstration[https://huggingface.co/spaces/swc2/Target-speaker-extraction?logs=container]
 +
** There are some channel issues, Poor performance when using some laptops or new launch Huawei phones
 +
** Guess it may be caused by some hardware denoise
 +
* BUPT host relocation
 +
** Still offline
 +
* Retraining for the best results in the paper experiment (2/10),and plan to release a version for IS2025 in the next two days at first
 
||
 
||
 
*
 
*
第74行： 第80行：
 
|Zehua Liu
 
|Zehua Liu
 
||
 
||
−
*
+
*Interspeech Paper Writing
 +
*Data checking
 +
*Code writing for loading LLM
 
||
 
||
 
*
 
*
第85行： 第93行：
 
|Pengqi Li
 
|Pengqi Li
 
||
 
||
−
*  
+
* IS 25(Fail temporarily and keep a record.)
 +
* Continue verifying the previous hypothesis
 
||
 
||
 
*
 
*
第96行： 第105行：
 
|Wan Lin
 
|Wan Lin
 
||
 
||
−
*  
+
* paper reading
 +
* NS experiments, but no new improvement now
 
||
 
||
 
*
 
*
第107行： 第117行：
 
|Tianhao Wang
 
|Tianhao Wang
 
||
 
||
−
*  
+
* code writing for the chain-based sound sep
 +
* two issue:
 +
** how to extract semantic emb related to corresponding label from mixture
 +
** similar label mapping (give to XiaoXue)
 
||
 
||
 
*
 
*
第118行： 第131行：
 
|Xiaoxue Luo
 
|Xiaoxue Luo
 
||
 
||
−
*  
+
* paper reading
 +
* code writing for the chain-based sound sep
 
||
 
||
 
*
 
*
第129行： 第143行：
 
|Zhenyu Zhou
 
|Zhenyu Zhou
 
||
 
||
−
*
+
* huawei project demo
 +
* dongxin project guide
 +
* some Personal matters
 
||
 
||
 
*
 
*
第140行： 第156行：
 
|Junhui Chen
 
|Junhui Chen
 
||
 
||
−
*
+
* Find a loss that may be helpful for NS[https://z1et6d3xtb.feishu.cn/docx/MxBNdPbLao0tsoxkBVCcUgUoneh#share-IpeLdu4PIodd3JxvjkccasOunJR], the code is ready.
 
||
 
||
 
*
 
*
第151行： 第167行：
 
|Jiaying Wang
 
|Jiaying Wang
 
||
 
||
−
*
+
* Confirm the implementation details of the baseline model（teacher forcing）
 +
* IS paper introduction part chinese version1（checking with zhenyu）[https://z1et6d3xtb.feishu.cn/docx/TUHldiaoQoYBqux7JEhcaCXenzh]
 
||
 
||
 
*
 
*
第173行： 第190行：
 
|Wenqiang Du
 
|Wenqiang Du
 
||
 
||
−
*  
+
* AI primary  handbook PPT （4/35）
 
||
 
||
 
*
 
*
第184行： 第201行：
 
|Yang Wei
 
|Yang Wei
 
||
 
||
−
*  
+
* Project
 +
** With the help of Qi Qu, finished zhongchuan ASR REST service APP.
 +
* Text Enroll KWS
 +
** Training on 500 hours data (use word level keyword sampling, add data from sinvxx)
 
||
 
||
 
*
 
*
第194行： 第214行：
 
|Turi
 
|Turi
 
||
 
||
−
*  
+
* Text collection for adding language model to Oromo ASR
 
||
 
||
 
*  
 
*  
第202行： 第222行：
 
|Yue Gu
 
|Yue Gu
 
||
 
||
−
*  
+
* Finegrained personality extractor is still in debug
 +
* Check the middle school handbook
 +
* Have been sick yesterday and today
 
||
 
||
 
*
 
*
第211行： 第233行：
 
|Qi Qu
 
|Qi Qu
 
||
 
||
−
*  
+
* KWS:
 +
** Pre-prod routine work for zh48+qingdao20 models.
 +
** Experiment on optimizing inference efficiency on Android devices failed: xnnpack acceleration does not work for dynamic shapes; quantization does not work either.
 +
* Misc:
 +
** C++/C libraries for KWS/AED processing released for mr536.
 
||
 
||
 
*
 
*

2025年1月20日 (一) 10:57的最后版本

People This Week Next Week Task Tracking (DeadLine)
Dong Wang
  • AI Graph high-school version, still on going
Lantian Li
Ying Shi
  • Huawei TSE online demo design with Zhenghai
  • Continue work on conditional overlap ASR model
Zhenghai You
  • TSE online service for Huawei project demonstration[1]
    • There are some channel issues, Poor performance when using some laptops or new launch Huawei phones
    • Guess it may be caused by some hardware denoise
  • BUPT host relocation
    • Still offline
  • Retraining for the best results in the paper experiment (2/10),and plan to release a version for IS2025 in the next two days at first
Junming Yuan
  • Go on MT-HuBERT paper writing(4/5)
  • Go on model pretraining
Xiaolou Li
Zehua Liu
  • Interspeech Paper Writing
  • Data checking
  • Code writing for loading LLM
Pengqi Li
  • IS 25(Fail temporarily and keep a record.)
  • Continue verifying the previous hypothesis
Wan Lin
  • paper reading
  • NS experiments, but no new improvement now
Tianhao Wang
  • code writing for the chain-based sound sep
  • two issue:
    • how to extract semantic emb related to corresponding label from mixture
    • similar label mapping (give to XiaoXue)
Xiaoxue Luo
  • paper reading
  • code writing for the chain-based sound sep
Zhenyu Zhou
  • huawei project demo
  • dongxin project guide
  • some Personal matters
Junhui Chen
  • Find a loss that may be helpful for NS[2], the code is ready.
Jiaying Wang
  • Confirm the implementation details of the baseline model（teacher forcing）
  • IS paper introduction part chinese version1（checking with zhenyu）[3]
Yu Zhang
  • Multi Agent Investment Result [4]
Wenqiang Du
  • AI primary handbook PPT （4/35）
Yang Wei
  • Project
    • With the help of Qi Qu, finished zhongchuan ASR REST service APP.
  • Text Enroll KWS
    • Training on 500 hours data (use word level keyword sampling, add data from sinvxx)
Turi
  • Text collection for adding language model to Oromo ASR
Yue Gu
  • Finegrained personality extractor is still in debug
  • Check the middle school handbook
  • Have been sick yesterday and today
Qi Qu
  • KWS:
    • Pre-prod routine work for zh48+qingdao20 models.
    • Experiment on optimizing inference efficiency on Android devices failed: xnnpack acceleration does not work for dynamic shapes; quantization does not work either.
  • Misc:
    • C++/C libraries for KWS/AED processing released for mr536.
取自“http://cslt.org/mediawiki/index.php?title=2025-01-20&oldid=42778”