“2026-04-20”版本间的差异

来自cslt Wiki
跳转至： 导航、 搜索
 
（14位用户的16个中间修订版本未显示）
第6行： 第6行：
 
|Dong Wang
 
|Dong Wang
 
||
 
||
−
*
+
* Check 2n semester textbook (done)
 +
* Refine ai textbook college version (1/2)
 +
* Check paper for micro-magenetics
 +
 
 
||
 
||
 
*
 
*
第28行： 第31行：
 
|Wenqiang Du
 
|Wenqiang Du
 
||
 
||
−
*
+
* Spech separation task
 +
** Reproduce the SpatialNet method (2 spk mix)
 +
** Train with mixed speakers: 2spk, 3spk, and 4spk，average SI-SDR up 7 (small data,incompelet separation)
 
||
 
||
 
*
 
*
第39行： 第44行：
 
|Yang Wei
 
|Yang Wei
 
||
 
||
−
*
+
* Audio separation model training with (L1 loss + SI-SDR loss) for low loudness problem
 
||
 
||
 
*
 
*
第61行： 第66行：
 
|Yue Gu
 
|Yue Gu
 
||
 
||
−
*
+
* write my Phd thesis（70% in progress）
 
||
 
||
 
*
 
*
第72行： 第77行：
 
|Lily
 
|Lily
 
||
 
||
−
*
+
* Question bank for three AI Literacy books
 +
* Lecturer Management System released
 +
* AIGE Center annual report PPT
 +
* AIGE routine work
 
||
 
||
 
*
 
*
第83行： 第91行：
 
|Pengqi Li
 
|Pengqi Li
 
||
 
||
−
*
+
* Core Work: Extended the findings of the ICASSP 2024 paper to the Chinese-language scenario
 +
* Progress Update: Drafted the paper for submission to Journal of Chinese Information Processing; the draft has been reviewed by Prof.Wang and is currently under revision.
 
||
 
||
 
*
 
*
第94行： 第103行：
 
|Junming Yuan
 
|Junming Yuan
 
||
 
||
−
*
+
* ZH paper refinement(done)
 
||
 
||
 
*
 
*
第105行： 第114行：
 
|Yu Zhang
 
|Yu Zhang
 
||
 
||
−
*
+
* GPU Util: https://z1et6d3xtb.feishu.cn/wiki/XX4NwX3tJiBDcgkMi0hcFUtInHh
 +
* Finish draft version of swarm paper with @chenjunhui
 +
* Continue running supplementary experiments (maybe three or four experiments still needed)
 
||
 
||
 
*
 
*
第116行： 第127行：
 
|Junhui Chen
 
|Junhui Chen
 
||
 
||
−
*
+
* get bad sick
 +
* paper refinement for several versions; references finding
 
||
 
||
 
*
 
*
第125行： 第137行：
  
 
|-
 
|-
−
|Jiaying Wang
+
|Xiaoxue Luo
 
||
 
||
−
*
+
* attractor-based USS
 +
** attractor counting accuracy of the 2-3mix model is still lower than expected
 +
** contact the author of the paper, will retrain based on his advice
 
||
 
||
 
*
 
*
第138行： 第152行：
 
|Bochao Hu
 
|Bochao Hu
 
||
 
||
−
*
+
* Topics-based VSR-LLM (100h topics data for now)
 +
** baseline: same encoder(FT) + transformer cer: 0.396 + Qwen-7B cer: 0.345 + Qwen-VL-7B cer: 0.318
 +
** + peripheral info is in training
 
||
 
||
 
*
 
*
第149行： 第165行：
 
|Hongcheng Zhang
 
|Hongcheng Zhang
 
||
 
||
−
*
+
*test qwen-omni model with different prompt
 
||
 
||
 
*
 
*
第160行： 第176行：
 
|Weiman Sun
 
|Weiman Sun
 
||
 
||
−
*
+
* Write my graduation thesis
 +
* Test the Qwen-omni models
 
||
 
||
 
*
 
*
第169行： 第186行：
 
||
 
||
 
*
 
*
 +
||
 +
*
 +
||
 +
*
 +
|-
 +
|-
 +
|Shuailong Li
 +
||
 +
*Reproduction on the MTASS original dataset（loss：MSE+SNR）
 +
**SDRi（speech=11.94 music=10.80 others=8.97 avg=10.57）
 +
**The perceived volume has not decreased
 +
*Reproduction on the MTASS original dataset（loss：MSE+SI-SDR）
 +
**Some problems have occurred and are being resolved
 
||
 
||
 
*
 
*

2026年4月20日 (一) 11:01的最后版本

People This Week Next Week Task Tracking (DeadLine)
Dong Wang
  • Check 2n semester textbook (done)
  • Refine ai textbook college version (1/2)
  • Check paper for micro-magenetics
Lantian Li
Wenqiang Du
  • Spech separation task
    • Reproduce the SpatialNet method (2 spk mix)
    • Train with mixed speakers: 2spk, 3spk, and 4spk，average SI-SDR up 7 (small data,incompelet separation)
Yang Wei
  • Audio separation model training with (L1 loss + SI-SDR loss) for low loudness problem
Ying Shi
Yue Gu
  • write my Phd thesis（70% in progress）
Lily
  • Question bank for three AI Literacy books
  • Lecturer Management System released
  • AIGE Center annual report PPT
  • AIGE routine work
Pengqi Li
  • Core Work: Extended the findings of the ICASSP 2024 paper to the Chinese-language scenario
  • Progress Update: Drafted the paper for submission to Journal of Chinese Information Processing; the draft has been reviewed by Prof.Wang and is currently under revision.
Junming Yuan
  • ZH paper refinement(done)
Yu Zhang
Junhui Chen
  • get bad sick
  • paper refinement for several versions; references finding
Xiaoxue Luo
  • attractor-based USS
    • attractor counting accuracy of the 2-3mix model is still lower than expected
    • contact the author of the paper, will retrain based on his advice
Bochao Hu
  • Topics-based VSR-LLM (100h topics data for now)
    • baseline: same encoder(FT) + transformer cer: 0.396 + Qwen-7B cer: 0.345 + Qwen-VL-7B cer: 0.318
    • + peripheral info is in training
Hongcheng Zhang
  • test qwen-omni model with different prompt
Weiman Sun
  • Write my graduation thesis
  • Test the Qwen-omni models
Ge Gao
Shuailong Li
  • Reproduction on the MTASS original dataset（loss：MSE+SNR）
    • SDRi（speech=11.94 music=10.80 others=8.97 avg=10.57）
    • The perceived volume has not decreased
  • Reproduction on the MTASS original dataset（loss：MSE+SI-SDR）
    • Some problems have occurred and are being resolved
取自“http://index.cslt.org/mediawiki/index.php?title=2026-04-20&oldid=44097”