“2024-05-27”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44行: 第44行:
 
|Zhenghai You
 
|Zhenghai You
 
||
 
||
*
+
* Revised dataset in correct SNR
 +
* Instance Normalization relieve data loudness differences
 
||
 
||
 
*
 
*
第81行: 第82行:
 
|Xiaolou Li
 
|Xiaolou Li
 
||
 
||
*
+
* Resnet3d frontend exp on LRS2
 +
* pos embed exp
 
||
 
||
 
*
 
*
第150行: 第152行:
 
|Junhui Chen
 
|Junhui Chen
 
||
 
||
*
+
* Neural Scoring complementary experiments
 
||
 
||
 
*
 
*

2024年5月28日 (二) 11:19的最后版本

People This Week Next Week Task Tracking (DeadLine)
Dong Wang
  • Several public talk (Henan, Aizhe, ChongQing, Kexie, Huawei)
  • NSFC review
  • Eletronics review
Lantian Li
Ying Shi
  • write CTC + fst decoding engine
    • CTC + fst TLG.fst decoding down
    • CTC + fst PLG.fst decoding in progress
Zhenghai You
  • Revised dataset in correct SNR
  • Instance Normalization relieve data loudness differences
Junming Yuan
  • make our Vlog and live broadcast video
  • prepare new research plan
Chen Chen
  • thesis defence
  • vii group
    • Structure: some experiment about 3d-cnn frontend with mamba on LRS2
    • Strategy: more experiments on audio memory (need adjust some hyper-parameters)
    • Data: 73 hours new, 612 hours in total
Xiaolou Li
  • Resnet3d frontend exp on LRS2
  • pos embed exp
Zehua Liu
  • AKVSR+pos_emb+attscore_CEloss(48.40) > AKVSR+pos_emb(48.80)[1]
  • NCMMSC2024 papper finish and i need check again
Pengqi Li
  • NC Papers
  • live broadcast
Wan Lin
  • NS: Complementary experiments
Tianhao Wang
  • Tracing experimental results
  • The influence of overlap ratio on EER threshold [2]
Zhenyu Zhou
Junhui Chen
  • Neural Scoring complementary experiments
Jiaying Wang
Yu Zhang
  • Paper reading for spatiotemporal time series forecasting and Financial related machine learning
Wenqiang Du
Yang Wei
  • Huilan stuff
    • Update TTS server for multiprocessing, to reduce latency.
Lily
Turi
  • Data collection
    • Audio checking for correctness, 26.5K so far.
  • Course works
Yue Gu
  • write paper: exp part 100%, method part 20%
  • complete Kespeech exps, performance meets expectations
  • find a new bug about semantic paraformer, retrain the model
Qi Qu
  • KWS
    • (ongoing) Standardize dataset formats and test routines.
    • Uyghur dataset: 129 speakers, 10 words.
    • New B6-based online model service.
    • TUI app for visualization.