VoxForge
VoxForge
Hi,
I am almost through with this tutorial but step 10 gives me this error.
HHEd
465/465 Models Loaded [5 states max, 1 mixes max]
RO 100.00 ''
Setting outlier threshold for clustering
RO->LS stats
and loading state occupation stats
ERROR [+7251] LoadStatsFile: unknown name b at line 1
FATAL ERROR - Terminating program C:\cygwin\HTK\htk-3.3-windows-binary\htk\HHEd
.exe
I can't figure out what the problem It complains of the 'stats' file generated in the previous step but the content of this file is the same as ones successfully cpmpleted.
Any help will be appreciated.
thanks.
Hi rovira74,
>ERROR [+7251] LoadStatsFile: unknown name b at line 1
From the HTK Book:
+7251 Stats ?le model errorFrom the HTK book:
A model name encountered in the stats ?le is invalid; check that the model set corresponds to that used to generate the stats ?le and that the stats ?le is complete and has not been corrupted.
For the ?nal pass of HERest [in Step 9], the -s option should be used to generate a ?le of state occupation statistics called stats. In combination with the means and variances, these enable likelihoods to be calculated for clusters of states and are needed during the state-clustering process described [in step 10].
The fulllist file contains the monophones from your entire pronunciation dictionary (not just the training pronunication dictionnary). Did you remember to merge the contents of fulllist and triphones1 into the new fulllist1 file?
It seems like your HHEd command can't find the first monophone (i.e. 'b') from your "fulllist" file in "fulllist1".
Hope that helps,
KenHi rovira74,
>ERROR [+7251] LoadStatsFile: unknown name b at line 1
In retrospect, I now think that the problem lies in the fact that the training prompts you created in Step 2 don't contain any words with the phone "b".
HHEd makes sure all monophones/biphones/triphones in your fulllist1 file are included in the hmmdefs file you create in Steps 9 - if it can't find one, it stops with an error. Try removing the "b" phone from your fulllist1 file to confirm this (there may be other phones that you need to remove).
You probably need to go back to Step 2 and add words that contain the phone 'b' (and any other phones that might be missing). In Step 10 you are creating triphones for the *entire* pronunciation lexicon (not just the training dictionnary you created from your grammar file)
In summary, you need to make sure you created a phonetically balanced prompt file - i.e that you included sentences with words (in addition to your grammar) that have all the phones in your entire pronunciation dictionary
Ken