“13-03-16 Mengyuan Zhao”版本间的差异

来自cslt Wiki
跳转至: 导航搜索
 
第11行: 第11行:
 
#Combine xEnt and mpe, test it.
 
#Combine xEnt and mpe, test it.
 
#Fix the problem in 1700H gmm training.
 
#Fix the problem in 1700H gmm training.
 +
#do mpe on sinovoice 1700H model using amr data(1700H)

2014年3月17日 (一) 02:41的最后版本

Weekly report:

  • Compare Fbank and GFbank feature.
  • Amr adaptation:
  1. trained xEnt and mpe, compare their perfomance.
  • Sinovoice:
  1. Convert 1700H wav into amr then convert back.
  2. Generate Fbank feature of 1700H corpus(amr).
  3. Trained gmm model on 1700H data.

Target for next week:

  1. Combine xEnt and mpe, test it.
  2. Fix the problem in 1700H gmm training.
  3. do mpe on sinovoice 1700H model using amr data(1700H)