|
|
| 第236行： |
第236行： |
| | |Qi Qu | | |Qi Qu |
| | || | | || |
| − | * | + | * KWS |
| | + | ** Standardize dataset formats and test routines. |
| | + | ** Data collection and processing. |
| | || | | || |
| | * | | * |
| People |
This Week |
Next Week |
Task Tracking (DeadLine)
|
| Dong Wang
|
|
|
|
| Lantian Li
|
|
|
|
| Ying Shi
|
- verify cohort Overlap ASR assumption
- Identify the speech component which most similar to the cohort vector ✔
- group work
|
- cohort + conditional chain Overlap ASR
|
|
| Zhenghai You
|
|
|
|
| Junming Yuan
|
- Continue to add various data augmentation functions into the code
- Prepare for live broadcast
|
|
|
| Chen Chen
|
|
|
|
| Xiaolou Li
|
- Video mamba exp (good good)
- patch frontend
- conv3d and resnet3d frontend
- Paper reading
|
- run exp on LRS2 and LRS3 (waiting for email feedback)
- what is the main difference between these two frontend? (conv3d and resnet3d)
|
|
| Zehua Liu
|
- AKVSR (cer:49.71%) > baseline(cer: 48.76%)
- AKVSR + pos_emb (a little worse)
- AKVSR + attention score loss(coding)
|
|
|
| Pengqi Li
|
- Jinfu and LiuHuan's Outlines of NC
|
- XueYing's Outline of NC
- NC paper of Speech XAI overview
|
|
| Wan Lin
|
|
|
|
| Tianhao Wang
|
- Baseline: SpEx+ with Detection (Failed)
- difficult to train because vox2 has a much larger data volume than wsj0
- Toolkit align: lr scheduler, pooling
- pooling seems critical (same epoch, NS loss: ASP is 0.16 vs TSP is 0.22)
|
|
|
| Zhenyu Zhou
|
|
|
|
| Junhui Chen
|
|
|
|
| Jiaying Wang
|
|
|
|
| Yu Zhang
|
|
|
|
| Wenqiang Du
|
|
|
|
| Yang Wei
|
|
|
|
| Lily
|
- PPT delivery
- Thesis
- Perception experiment
|
|
|
| Turi
|
- Data Collection
- Class works
|
|
|
| Yue Gu
|
|
|
|
| Qi Qu
|
- KWS
- Standardize dataset formats and test routines.
- Data collection and processing.
|
|
|
