VoxForge
VoxForge
Hi,
when I execute the HHEd command to create the files in the hmm10 folder I get this warning message:
HHEd -A -D -T 1 -H hmm9/macros -H hmm9/hmmdefs -M hmm10 mktri.hed monophones1
...
TI T_sil {}
Tie items
WARNING [-2631] ApplyTie: Macro T_sil has nothing to tie of type t in prog/HHEd
Saving new HMM files ...
Edit Complete
No HTK Configuration Parameters Set
Can anybody help me with this problem? I can't find my mistake! Regards!
Ok, I think I found the answer to this problem in the HTK-Book! The Book says to this warning: "When running the HHEd command you will get warnings about trying to tie transition matrices for the sil an sp models. Since neither model is context-dependet there aren't actually any matrices to tie".
So, with other words: Don't worry about this warning massage!
But I got another problem, when I execute the HERest command to create the files in the hmm11 folder, then I get these warning massages:
...
WARNING [-2331] UpdateModels: a-R+f[2665] copied: only 1 egs
in prog/HERest
WARNING [-2331] UpdateModels: a-R+k[2666] copied: only 1 egs
in prog/HERest
WARNING [-2331] UpdateModels: R-ih+ah[2668] copied: only 1 egs
in prog/HERest
WARNING [-2331] UpdateModels: a-R+p[2670] copied: only 1 egs
in prog/HERest
Total 173 floored variance elements in 142 different mixes
Saving hmm's to dir hmm11
Reestimation complete - average log prob per frame = -5.337948e+01
- total frames seen = 7.870700e+04
Can anybody tell me why I get these warnings?
Regards, Moe
Hi Moe,
>WARNING [-2331] UpdateModels: a-R+p[2670] copied: only 1 egs
I think this one basically means you need more audio data for the triphone to be properly trained. The HTK book or the HTK email archives will have more info on this.
Ken