MGE方法（DJ）的几个注意

理论部分参考Paper

Y. J. Wu and R. H. Wang, "Minimum generation error training
for HMM-based speech synthesis," in Proc. of ICASSP 2006, vol.
1, pp. 89-92, May. 2006.
Y. J. Wu,　R. H. Wang　and F. Soong, "Full Hmm training for minimizing generation error in synthesis," in Proc. of ICASSP 2007

实际操作中，注意

1、 forcealignment生成的full_a.mlf，可能用中间结果的fullcontext

 HVite -l '*' -A -a -B  -C trn.cnf -H fullcontext.mmf -i full_a.mlf  -m -y lab -I full.mlf -S train.scp full.list_d full.list

2  去掉 -B  生成 text版model 

 

未完待续