“2024-06-03”版本间的差异

• Paper for Uyghur children pronunciation analysis (@Maolidan)
• Review NSFC
• Several public talks

• GPU status [1]
• Projects

- AED -> Quantization - TSE -> SpkAug, InstanceNorm, AbsentLoss - VSR -> reaching 1000h - Finance -> Few-shot, RAG + LLM

• Papers

- NeuralScoring -> Paper-Exps, Speed up the progress

• Postgraduate thesis & defense

• Finish phone-level CTC decoding with TLG fst
• Some stuff about Huawei project collaboration

• Restart research about Cohort Overlap ASR

• Revised dataset with correct snr
• Train targetless data with SE-SISDR loss[2]

• Adjusted the strategy of "learn not to listen", but still no work[3]
• make the plan for the next stage

• Pos embedding Exp [4]
• Frontend and bi-Transformer Exp
• Branchmamba Exp

• Smooth Label Celoss (cer: 48.09)[5]
• Find some parameters need to adjust[6]

• Completed the NC paper(v1) titled "Explainable Artificial Intelligence for Deep Speech Signal Processing: A survey"

• Supplement experimental results
  • NS & Mixup & Eaasp single/mutil-scenario training

• Tracing experimental results
  • "hard spk sample" is useful for clean ns
• share encoder exps (training)

• huawei project[7]

• Neural Scoring complementary experiments

• come back! continue cohort within vit structure

• Paper rading
• Huawei cough humming model quantization

• Prepare to train new kws model （48 words） for Aibabel Company

• Huilan TTS
  • Fix bugs of the updated TTS server.
  • Try to use ONNX model in TTS server. (In progress)

• Live broadcast
• Xinjiang teacher's course
• Children's course (Sunday)

• Data collection
  • No contribution, as students are on final
• Final projects and live broadcast

• complete the semantic paraformer exps
• do some modular analysis about Competition Elimination Mechanism

• KWS
  • Data processing routines: VAD, ASR.
  • Test routines: keyword-wise ROC.
  • Data collection: Uyghur and Kazakh.
• AED
  • Online demo.