|
|
| 第22行： |
第22行： |
| | * GPU status [https://z1et6d3xtb.feishu.cn/wiki/XGcGwRK5viJmpRkjH9AczIhynCh] | | * GPU status [https://z1et6d3xtb.feishu.cn/wiki/XGcGwRK5viJmpRkjH9AczIhynCh] |
| | * Rebutal for IJCAI paper | | * Rebutal for IJCAI paper |
| − | * ASIP-BUPT (CohortTSE, SE-Adapter, SpeakerAug, NeuralScoring) | + | * ASIP-BUPT (CohortTSE, NeuralScoring) |
| | || | | || |
| | * | | * |
| People |
This Week |
Next Week |
Task Tracking (DeadLine)
|
| Dong Wang
|
- Design learn not to listen, to reduce false alarms for KWS [1]
- Design AI courses for primary and middle schools
- Finaly review of NeuralMag paper [2]
- Rebutal for IJCAI paper [3]
|
|
|
| Lantian Li
|
- GPU status [4]
- Rebutal for IJCAI paper
- ASIP-BUPT (CohortTSE, NeuralScoring)
|
|
|
| Ying Shi
|
- Experimental Result for SPL Paper: Phrase Guided End-to-End Target Sentence Extraction from Overlapping Speech
- Data Preparation
- Chinese Overlap ASR model
- Detect wake-up words from Continuous speech
|
- Finish draft for SPL paper
- Finish model training
|
|
| Zhenghai You
|
- A matching encoder experiment in cohort[5]
|
|
|
| Junming Yuan
|
- Experimental report on "learn not to listen"[6]
|
|
|
| Chen Chen
|
- entropy analysis
- group work [7]
|
|
|
| Xiaolou Li
|
- Reproduce different structure
- Baseline training with less data
- Code write and debug
- ResNet3D, Branchformer, E-Branchformer, interCTC
- Paper Reading: Some VSR paper in ICASSP2024
|
- Experiment in different structure
|
|
| Zehua Liu
|
- ASR training for model distillation
- Paper Reading
|
|
|
| Pengqi Li
|
- summary[8] of speech processing XAI for NSFC
- download 160+ papers(113 about speech processing XAI; Traditional XAI method; review)
- Summarize them using LLM and categorize by speech processing task.
|
|
|
| Wan Lin
|
|
|
|
| Tianhao Wang
|
- Neural scoring docs and codes reviewing
|
|
|
| Zhenyu Zhou
|
- Neural scoring docs and codes reviewing
|
|
|
| Junhui Chen
|
- Neural scoring: mix/overlap/concat test
|
|
|
| Jiaying Wang
|
- one cohort distance test(The probability of selecting the source closer to the cohort as the target during testing)
- the rate is still around 0.5
- speakerbeam with no enroll/cohort + minimal loss training
- double-check done
- still training, but seem to converge at val_loss around -3
- confused
|
|
|
| Yu Zhang
|
- Financial Backtest indicators check
- Jun Wang R2 SAC codes and papers reading
|
- Reproduce R2 SAC and FinRL policy
|
|
| Wenqiang Du
|
- update CN KWS model for AIbabel
- Using real environment and FA data to update the model
|
|
|
| Yang Wei
|
- Prepare for children mispronunciation detection and diagnosis base model
|
|
|
| Lily
|
- Paper reading and prepare for journal paper
- Data annotation (for perception)
|
|
|
| Turi
|
- Data Collection App Backend [10]
- User authentication
- Data storage
|
|
|
