| People |
This Week |
Next Week |
Task Tracking (DeadLine)
|
| Dong Wang
|
- Presentation for several public AI promotion
- Primary AI, Grade 3 (1), 8 chapters done
- Some review on linguistic literature about perception uncertainty in pronunciation assessment
- New form of text enrollment and speech fine-tuning to account for accent-based KWS.
|
|
|
| Lantian Li
|
- GPU status [1]
- Projects (AED, TSE)
- ASIP-BUPT (NeuralScoring, CohortSS)
- BlockChain Courses
|
|
|
| Ying Shi
|
- DI-TING structure verify here
|
|
|
| Zhenghai You
|
|
|
|
| Junming Yuan
|
- AI Graph slides refinement
- IS24 rebuttal
- Control FA experiment baseline result[2]
|
|
|
| Chen Chen
|
- prepare CNVSRC2024 Baseline system
- vii group [3]
|
|
|
| Xiaolou Li
|
- CNVSRC2024 baseline training
- IS24 Rebuttal
|
|
|
| Zehua Liu
|
- AKVSR code
- data crop
- IS24 Rebutall
|
|
|
| Pengqi Li
|
|
- Read&Summary SA-XAI workshop(ICASSP) papers
- Experiment(PID) on Timit(Extend workshop paper)
|
|
| Wan Lin
|
- Pre & QA for graduation paper
- Neural Scoring [4]
|
|
|
| Tianhao Wang
|
- Neural Scoring [5]
- IS24 rebuttal
|
|
|
| Zhenyu Zhou
|
- Paper reading
- IS24 rebuttal
|
|
|
| Junhui Chen
|
- Graduation paper
- Neural Scoring: chunk 2s->4s, NS is better than EA-ASP
|
- Try to use large pretrain model as test utt encoder(wavLM, wav2vec2, etc.)
|
|
| Jiaying Wang
|
- paper reading report
- Huawei data collection
- cohort test[6]
|
|
|
| Yu Zhang
|
- AutoML:
- switched from FLAML to EVALML as it provided a tool chain better suited to our tasks
|
|
|
| Wenqiang Du
|
- Hard negative training,FA data is ready,prepare to train the model
-
|
|
|
| Yang Wei
|
- Children mispronunciation detection and diagnosis
- Prepare baseline recipe and challenge document
- Check the baseline model, due to the extraordinary CER performance.
|
|
|
| Lily
|
|
|
|
| Turi
|
- Data collection
- Course work & paper reading
|
|
|
| Yue Gu
|
- Semantic paraformer model reconstruction
- Interspeech rebuttal
|
|
|
| Qi Qu
|
- Performance test:
- Two-phased KWS vs KWS + FunASR
- Data processing:
- ~100k FA segments collected out of ~6k hours
- Data processing routine
- Model training:
- Fine-tuning w/ ~20k more FA
|
|
|
