I think (gamma=0, alpha=0) gives cep, other than lsp.
BTW, the silent problem was discussed in this thread:http://hts.sp.nitech.ac.jp/hts-users/spool/2012/msg00325.html
There is no solution yet, but it might help shooting.
Beijing Institute of Technology [naxy(at)bit.edu.cn]
Idiap Research Institute [xingyu.na(at)idiap.ch]
At 2012-11-29 14:18:49,ArthurLeo <bin007.zhao@xxxxxxxxxxx> wrote:
>HTS2.2 straight demo used lsp(gamma=0,alpha=0), the syn files are silent. I checked the parameter before training,the parameters after mgcep converting(sp to lpc), I guess it maybe the lpc, it is very small, especially the first in every frame, like 0.000100683, but lpc got from the common demo(not using straight sp coverting) is not so small, it is 2.26152 corresponding to the above data.
>Is it normal that lpc is so small using straight demo？