2014年12月28日-SGR
来自cslt Wiki
1. speech transformed to image, classification
2. [pass] MFCC+
3. EGG usage. insertion is not important
4. co-training. GMM-HMM adaption. class-based LM adaptation.
5. [conflict] CNN.tone information. F0, FFV.
6. [important] spatial histogram equalization. MAS-HEQ
7. blind spatial subtraction.
8. denoising DAE. multi-speaker.
