“2025-04-14”版本间的差异

来自cslt Wiki
跳转至： 导航、 搜索
 
（2位用户的2个中间修订版本未显示）
第104行： 第104行：
 
|Wan Lin
 
|Wan Lin
 
||
 
||
−
*  
+
* Some Experiments [https://z1et6d3xtb.feishu.cn/docx/MxBNdPbLao0tsoxkBVCcUgUoneh?from=from_copylink]
 
||
 
||
 
*
 
*
第115行： 第115行：
 
|Tianhao Wang
 
|Tianhao Wang
 
||
 
||
−
*  
+
* huawei project & huohua testing
 +
* personal stuff
 
||
 
||
 
*
 
*
第163行： 第164行：
 
* explore Whisper: ASR function cannot transcribe mix speech well, the SOT of whisper is on the task level (translate, transcribe, ...)
 
* explore Whisper: ASR function cannot transcribe mix speech well, the SOT of whisper is on the task level (translate, transcribe, ...)
 
* modified the model
 
* modified the model
−
** using the CTC loss with the minimum loss per iteration backpropagation strategy: loss still turns out to be NaN
+
** using the CTC loss with the minimum loss per iteration backpropagation strategy: loss still turns out to be NaN (problem not in loss caculation)
 
** replace asr modules to find reason
 
** replace asr modules to find reason
 
* other ASR tools
 
* other ASR tools

2025年4月14日 (一) 11:05的最后版本

People This Week Next Week Task Tracking (DeadLine)
Dong Wang
  • AIGE opening ceremoney
Lantian Li
  • Complete IS rebuttal
  • Translate ML book
  • Release AIGE web v.10
Ying Shi
  • 4.11 meeting stuff
  • Structure of thesis
Zhenghai You
  • Some model tuning in Huawei projects (training)
  • Revise rebuttal
Junming Yuan
  • make primary school PPT
  • some AI practice handbook works
Xiaolou Li
  • Write document for GA project
  • LipVoicer reproduction[1]
Zehua Liu
  • Reproduce ACClip2Speech on CNVSRC.Single[2]
  • Finish Vocoder code (still on training)
  • GongAn document writing
Pengqi Li
  • developed primary school PPTs (2)
  • Co-developed AI practice handbooks with the practice schools (3)
Wan Lin
  • Some Experiments [3]
Tianhao Wang
  • huawei project & huohua testing
  • personal stuff
Xiaoxue Luo
  • Sound separation
    • baseline: adjust the code of one USS model and evaluate its performance using our test datasets
Zhenyu Zhou
Junhui Chen
  • Continue all NS exp with 1 layer
  • Read paper
Jiaying Wang
  • explore Whisper: ASR function cannot transcribe mix speech well, the SOT of whisper is on the task level (translate, transcribe, ...)
  • modified the model
    • using the CTC loss with the minimum loss per iteration backpropagation strategy: loss still turns out to be NaN (problem not in loss caculation)
    • replace asr modules to find reason
  • other ASR tools
    • trying mix speech ASR of FunASR, and attempting to integrate it into the current model
Yu Zhang
  • Add additional inference step for technical factor analysis (worth for a longer backtesting)
  • Done single stock trade pipeline (debug is done, can run backtesting tomorrow)
  • Weight avg trading (still debugging)
Wenqiang Du
  • Prepared for the 4.11 meeting
Yang Wei
  • Some modification based on the kws patent document from Wenqiang.
Turi
Yue Gu
  • read the FIP paper carefully， write two introduction paragraphs
  • reproduce some hotword exps in new cloud service
Qi Qu
  • Finished organizing codebase for evaluating KWS models (both "audio-enroll" and text-enroll).
  • Text-enroll KWS models evaluation on Mandarin Chinese w/ southern China accent data [4].
取自“http://index.cslt.org/mediawiki/index.php?title=2025-04-14&oldid=43030”