See this post: error 7321

but i do not want to get hmm14,i want to get hmm11, hmm12 and stats.I think i don't need to add the monophone0 to triphones1 file.

But when i add the monophone to the triphones1,the error 7321 is being solved .

why?

Hi xuexiaoyan,

>>CreateInsts: Unknown label sh

>But when i add the monophone to the triphones1,the error 7321 is

>being solved .  why?

Because the 'sh' label is in your triphones file and not in your monophones file.

When you create your fulllist1 file in Step 10, you create a phone list against the *entire* pronunciation dictionary - so it includes the 'sh' phone.

I think you are using a grammar that is different than your training prompts.  One or more words in your grammar file use the 'sh' prompt, but you did not record any audio for those words, therefore no monophone was created.

In Step 10, you create "state-tied triphones" which creates logical triphones from your *entire* pronunciation lexicon to the actual physical triphones in your acoustic model.  It does this by linking an unseen triphone (one that is in your pronunciation dictionary, but for which you have not provided any training data) with triphones in your training set.  The logical triphones with 'sh' in them must be "tied" to some other physical triphone.  You should be able to confirm this by looking at your tiedlist file.

Hope that helps,

Ken

the command that i used is "

herest -T 1 -C config -s trainlist/train.scp -H models/hmm10/macros -H models/hmm0/hmmdefs -M 250.0 150.0 1000.0  -s stats triphonetrain.mlf triphone1"

the result that i got is "

pruning -on[250.0 150.0 1000.0]

...

processing data :DR1FDML0SX159.mfc ; label DR1FDML0SX159.lab utterance prob per frame=-7.062950e+001

processing data :DR1FDML0SX249.mfc, label DR1FDML0SX249.lab

ERROR [+7321]  CreateInsts: Unknown label ey

 FATAL ERROR - Terminating program HERest"

i exame every file ,expecially DR1FDML0SX249.lab ,i can't find the monophone "ey".

i don't get "stats" file,so i can't do the next steps which include generating the tiedlist and fullist.

Hi xuexiaoyan,

>herest -T 1 -C config -s trainlist/train.scp -H models/hmm10/macros

>-H models/hmm0/hmmdefs -M 250.0 150.0 1000.0  -s stats

>triphonetrain.mlf triphone1"

I am assuming you mean that you executed something like this:

HERest  -A -D -T 1 -C config -I wintri.mlf -t 250.0 150.0 3000.0

-s stats -S train.scp -H hmm11/macros -H hmm11/hmmdefs

-M hmm12 triphones1

>i exame every file ,expecially DR1FDML0SX249.lab ,i can't find the

>monophone "ey".

Now that is interesting...

Is there an 'ey' phone in your hmm9/hmmdefs file? 

Is there an 'ey' phone in your monophones1 file?  Since it is used to create your mktri.hed file... is there an 'eh' phone in your mktri.hed file?

Might be easier to restart the steps from you last good run of HERest...

Ken