“2024-05-20”版本间的差异

来自cslt Wiki
跳转至: 导航搜索
 
(6位用户的6个中间修订版本未显示)
第23行: 第23行:
 
||
 
||
 
* GPU status [https://z1et6d3xtb.feishu.cn/wiki/XGcGwRK5viJmpRkjH9AczIhynCh]
 
* GPU status [https://z1et6d3xtb.feishu.cn/wiki/XGcGwRK5viJmpRkjH9AczIhynCh]
* Projects (AED -> Hardware support, TSE -> Test&Analysis)
+
* Projects (AED -> Quantization, TSE -> Data analysis & Model training)
* ASIP-BUPT (NeuralScoring -> Paper, CohortSS -> Data Analysis)
+
* ASIP-BUPT (NeuralScoring -> Paper-Exps, CohortSS -> Hold)
* Check NIPS & Review theses
+
* My application & HaiNan fund
 
||
 
||
 
*  
 
*  
第36行: 第36行:
 
|Ying Shi
 
|Ying Shi
 
||
 
||
*
+
* Diting Demo
 +
* Cohort ASR with conditional chain
 +
* [https://z1et6d3xtb.feishu.cn/docx/Mt7adFRZMoDHcwxkeLPcdWTGngf?from=from_copylink group work]
 
||
 
||
 
*
 
*
第49行: 第51行:
 
* Data extension model in session one data
 
* Data extension model in session one data
 
* Evaluation of mixed speech of targetless speakers
 
* Evaluation of mixed speech of targetless speakers
* Model evaluation under different loudness levels
+
* Model evaluation under different loudness levels[https://z1et6d3xtb.feishu.cn/docx/PBAZdsiSWoq82YxWsu3cCD4Tnte]
 
||
 
||
 
*
 
*
第101行: 第103行:
 
* AKVSR + pos_emb + attscore_celoss very promising
 
* AKVSR + pos_emb + attscore_celoss very promising
 
* AKVSR(v2.1) + pos_emb  > AKVSR + pos_emb
 
* AKVSR(v2.1) + pos_emb  > AKVSR + pos_emb
* NC2024 paper writing  
+
* NC2024 paper writing [https://z1et6d3xtb.feishu.cn/docx/ZaTFd3A5EoK982xWBVschloanee?from=from_copylink]
 
||
 
||
 
*
 
*
第123行: 第125行:
 
|Wan Lin
 
|Wan Lin
 
||
 
||
*
+
* NS: add enroll condition to test encoder: didn't work
 +
** Mix-Vox1-O: 4.94% vs 4.28%(vit)
 +
* Determine the experimental design of NS paper together
 
||
 
||
 
*
 
*
第193行: 第197行:
 
|Wenqiang Du
 
|Wenqiang Du
 
||
 
||
* Upgrade task for aibabel's new 48 worlds KWS model,Doing
+
* Upgrade task for aibabel's new 48 words KWS model,Doing
 
||
 
||
 
*
 
*
第237行: 第241行:
 
|Yue Gu
 
|Yue Gu
 
||
 
||
*
+
* paper writing: 70% experiment section
 +
* compelete kespeech baseline
 +
* the training of semantic paraformer is convergent
 
||
 
||
 
*
 
*

2024年5月21日 (二) 01:02的最后版本

People This Week Next Week Task Tracking (DeadLine)
Dong Wang
  • Prepare several public presentations
Tsinghua middle school - Yongfeng
Tsinghua Middle school - Zhixin
BanYueTan & Xinhua pub. & Tsinghua pub.
  • Text-to-sql paper polishing
Lantian Li
  • GPU status [1]
  • Projects (AED -> Quantization, TSE -> Data analysis & Model training)
  • ASIP-BUPT (NeuralScoring -> Paper-Exps, CohortSS -> Hold)
  • My application & HaiNan fund
Ying Shi
  • Diting Demo
  • Cohort ASR with conditional chain
  • group work
Zhenghai You
  • Data extension model in session one data
  • Evaluation of mixed speech of targetless speakers
  • Model evaluation under different loudness levels[2]
Junming Yuan
  • Add speed perturbation into the code, still no improvement.
  • Control FA experiment result[3]
  • To design other method
Chen Chen
  • Prepare for live broadcast
  • vii group [4]
    • good news for audio memory and bi-mamba
    • 87 hours new data for CN-CVS2
Xiaolou Li
  • Bi-Mamba exp on cnvsrc-single (48.75%)
  • LRS2, LRS3 dataset preparation
  • Bi -Mamba exp on LRS2 (4s and full, slow training)
  • try to find a method to do data cleaning
Zehua Liu
  • AKVSR + pos_emb > baseline
  • AKVSR + pos_emb + attscore_celoss very promising
  • AKVSR(v2.1) + pos_emb > AKVSR + pos_emb
  • NC2024 paper writing [5]
Pengqi Li
  • NC papers of Our Group[6]
Wan Lin
  • NS: add enroll condition to test encoder: didn't work
    • Mix-Vox1-O: 4.94% vs 4.28%(vit)
  • Determine the experimental design of NS paper together
Tianhao Wang
  • align the toolkit for neural scoring
  • next experimental designing with CJH and LW
Zhenyu Zhou
Junhui Chen
Jiaying Wang
Yu Zhang
  • AutoML
    • slash stock num
    • last 7 days to forecast next 1 day return value (failed)
    • last 7 days to forecast next 1 day rise or fall (failed)
Wenqiang Du
  • Upgrade task for aibabel's new 48 words KWS model,Doing
Yang Wei
  • Children MDD challenge
    • Discuss with CASS, and prepare environment on their device
  • Huilan stuff
    • try to convert the TTS model to ONNX format
Lily
  • Thesis
Turi
  • Data Collection
    • 25.9K
  • Paper reading and Course Works
Yue Gu
  • paper writing: 70% experiment section
  • compelete kespeech baseline
  • the training of semantic paraformer is convergent
Qi Qu
  • KWS
    • (ongoing) Standardize dataset formats and test routines.
    • Optimize B6-based online model service.