“2024-01-29”版本间的差异

• MicroMagnetic paper, the first pass completed.

• GPU status [1]
• ASIP-BUPT (Neural Scoring)
• ASIP Annual report

• Keyword-Attributed OverLap ASR
  • Fix test dataset: LibirMix-Espnet & LibriMix-Official (2 mix clean)
  • Finish model training: KA-ASR-Full, KA-ASR-Oracle, SOT-Our
• Cohort Overlap ASR
  • Finish first step: Recognize one source from mixture by employ speaker embedding

• Check and organize the mix-training pretraining experiment project.

·Solving the error of MFA on dragon03.(done)
·Extending the pretraining data.(done)
·Exploring the effect of BN in the few-shot finetuning(in progress).

• CNCVS data collect
  • Finished testing phase with support from sunyiwei,shuyanzhi,mengshuaiming
• Child Record Website
  • Finished phoneme annotation phase
  • Get some statistics
• DeepFake
  • Human Test on DFDC
  • Zehua & Xiaolou Report

• Signal leval Speaker Augmentation Plan[2]:
  • Transformation(Random based & Knowledge based)
  • Speaker Characteristics Guided Voice Conversion

• speaker encoder preparation（ResNet34_ASP_AAMSoftmax-LMFT）

• gender divide test on speaker beam
• cohort with min SNR loss pb

• Diting Projiect
  • data aug
  • add gaussian noise to control FA

• Huilan stuff
  • Develop stream mode ASR interface for ASR service
  • Deal with time delay problem with long text input for TTS service

• update statistical results[3]