Acoustic Model Discussions

Flat
Step 2 error
User: Jozo
Date: 2/19/2008 11:42 am
Views: 7174
Rating: 21

I am trying to create acoustic model using auto script but I am getting very strange error in step 2.

It says something like error in HDMan, cat: ./interim_files/monophones1, no such file or directory

and then in the next line something like 

 cat:logs/Step2_HDMan.log no such file or directory

 

It looks like there is some problem with cat command but I

can`t figure out what and how it is connected to HDMan. And why

there is no monophones1 file when HDMan is supposed to create

it. I have 10 words (non english) in my dictionary and I added 

them all to voxforge lexicon at the begining of file. 

--- (Edited on 2/19/2008 11:42 am [GMT-0600] by Visitor) ---

Re: Step 2 error
User: kmaclean
Date: 2/19/2008 1:29 pm
Views: 237
Rating: 18

Hi Jozo,

I really need to see the exact errors to be able to help you out. 

I am assuming you are running the "HTK_Compile_Model.sh" script from the how-to.  The relevant code from this script is as follows:

print_heading "Step 2 - Pronunciation Dictionnary"
make_wlist_label_file
make_lexicon
HDMan -A -D -T 1 -m -w ./interim_files/wlist -e ./input_files -n ./interim_files/monophones1 -i -l logs/Step2_HDMan_log ./interim_files/dict ../../lexicon/voxforge_lexicon > logs/Step2_HDMan.log
make_monophones0
echo -e '***Please review the following HDMan output***:\n'
cat logs/Step2_HDMan_log

Therefore the reason why the cat command is failing is that it can't find the file it is supposed to display to the console.  This is likely caused by problems in the execution of HDMan.

> I have 10 words (non english) in my dictionary and I added them all to

>voxforge lexicon at the begining of file.

HDMan assumes that the dictionary is sorted ... so adding new words to the *beginning* could be the cause of the problem.

In addition, the HTK_Compile_Model.sh script assume English.  You could probably modify it to compile for non-English languages, but then you need to look at creating an HTK tree.hed script for that language.

You might have better luck following the VoxForge Tutorial and creating monophone models (Step 8 in the tutorial), or triphone models (Step 9).  

To be able to complete Step 10 and create tied-state acoustic models you need a tree.hed script for your language.  For more information on how to create a tree.hed file for a new language, see the following links:

Ken 

--- (Edited on 2/19/2008 2:29 pm [GMT-0500] by kmaclean) ---

Re: Step 2 error
User: Speerec
Date: 2/20/2008 6:07 am
Views: 256
Rating: 20

HDMan command creates wlist (./interim_files/wlist) normally.

>HDMan assumes that the dictionary is sorted ... so adding new >words to the *beginning* could be the cause of the problem. In >addition, the HTK_Compile_Model.sh script assume English.

Does this really have any difference? Because, we use english phonemes to "describe" words. Other thing, we moved our new words from file, leaving default dictionary, and it doesn't work either.

When i made manualy monophones1 file(empty),  HDMan commadn deletes it. The Step2_HDMan_log is created, but its empty.

 

--- (Edited on 2/20/2008 6:07 am [GMT-0600] by Visitor) ---

Re: Step 2 error
User: kmaclean
Date: 2/20/2008 7:47 pm
Views: 218
Rating: 22

>HDMan command creates wlist (./interim_files/wlist) normally.

HDMan does not create the wlist file, it uses it to create the dict and monophones1 files.

Try following Step 2 of the Tutorial and see if that helps.  The tutorial basically does everything the script does, but manually, and with explanations.

Ken 

--- (Edited on 2/20/2008 8:47 pm [GMT-0500] by kmaclean) ---

Re: Step 2 error
User: Jozo
Date: 2/21/2008 4:40 am
Views: 271
Rating: 20

Here is the error:

 

-e      init
=====================================

-e      Step 1 - Task Grammar
=====================================

-e already completed manually


-e      Step 2 - Pronunciation Dictionnary
=====================================
sorting:./interim_files/wlist to:./interim_files/wlist1
Found voxforge_lexicon
./HTK_Compile_Model.sh: 225: HDMan: not found
cat: ./interim_files/monophones1: No such file or directory
-e ***Please review the following HDMan output***:

cat: logs/Step2_HDMan_log: No such file or directory
 

--- (Edited on 2/21/2008 4:40 am [GMT-0600] by Visitor) ---

Re: Step 2 error
User: kmaclean
Date: 2/21/2008 7:29 am
Views: 2986
Rating: 18

Hi Jozo,

>./HTK_Compile_Model.sh: 225: HDMan: not found

This means that the acoustic model creation script cannot find HTK's HDMan command.  Please make sure HTK was compiled correctly, and that you set your path correctly - see this link.

Can you run HDMan from the command line? 

Ken 

--- (Edited on 2/21/2008 8:29 am [GMT-0500] by kmaclean) ---

PreviousNext