Acoustic Model Discussions

Nested
Speech Recognition Model Converter
User: kmaclean
Date: 5/3/2007 10:48 am
Views: 13823
Rating: 40

Found a project on sourceforge that has code for a universal speech recognition model converter (USMC):

    Speech Recognition Model Converter

The project admins seem to Arthur Chan (creator of the article Why there is no Open Source Dictation) and Udhyakumar Nallasamy - who gave us some info on the use of mp3 audio for AM creation in this post).

Udhyakumar Nallasamy home page states that:

This tool can convert sphinx continuous models into HTK compatible models. The present version of the code can be obtained from the CVS of sourceforge. 
 

Ken 



--- (Edited on 5/3/2007 11:48 am [GMT-0400] by kmaclean) ---

Re: Speech Recognition Model Converter
User: dara
Date: 1/12/2011 11:25 am
Views: 183
Rating: 17

Hi,

First of all, sorry for replying an old post. I was looking for a way to convert acoustic model format from Sphinx to HTK and found your post. I clicked the link, but there's no file at all on Speech Recognition Model Converter. The author might have removed it. Is there other resources useful to perform that task? Thanks in advance.

regards, Dara

--- (Edited on 1/12/2011 11:25 am [GMT-0600] by dara) ---

Re: Speech Recognition Model Converter
User: kmaclean
Date: 1/12/2011 5:07 pm
Views: 172
Rating: 16

>I clicked the link, but there's no file at all on Speech Recognition Model

>Converter.

You need to click the Develop tab and click browse code to get this link: here: http://srmc.cvs.sourceforge.net/viewvc/srmc/usmc/

--- (Edited on 1/12/2011 6:07 pm [GMT-0500] by kmaclean) ---

Re: Speech Recognition Model Converter
User: dara
Date: 1/13/2011 1:09 am
Views: 155
Rating: 15

Didn't notice that. Thanks a million, kmaclean.


regards,

Dara

--- (Edited on 1/13/2011 1:09 am [GMT-0600] by dara) ---

Re: Speech Recognition Model Converter
User: enigma1987
Date: 1/14/2011 3:25 am
Views: 142
Rating: 15

Hi Dara,

I also need to convert sphinx 3 and pocketsphinx acoustic models to HTK acoustic model and did not find a way to do the conversion.

Did you manage to use the converter ?

I also downloaded the srmc but I could not use it on Linux (Ubuntu).

 

--- (Edited on 1/14/2011 3:25 am [GMT-0600] by enigma1987) ---

Re: Speech Recognition Model Converter
User: dara
Date: 1/15/2011 11:35 am
Views: 160
Rating: 14

Hi enigma1987,


Yes, I did. I built the SRMC with Visual C++ 2008 on Windows. All you need to do is open the project file, and build the application from it. I don't know how to use it on Linux, sorry.

 

regards,

Dara

--- (Edited on 1/15/2011 11:35 am [GMT-0600] by dara) ---

Re: Speech Recognition Model Converter
User: enigma1987
Date: 1/15/2011 5:53 pm
Views: 121
Rating: 14

Hi,

I setup visual c++ 2008 and when I try to build SphinxTrain project it gives 3 errors:

 

4>libcommon.lib(cmd_ln.obj) : error LNK2005: _cmd_ln_parse already defined in s3_libutil.lib(cmd_ln.obj)

4>libcommon.lib(cmd_ln.obj) : error LNK2005: _cmd_ln_access already defined in s3_libutil.lib(cmd_ln.obj)

4>.\..\..\..\..\bin\Debug/s3mod2htk.exe : fatal error LNK1169: one or more multiply defined symbols found

 

 

When I try to build only s3mod2htk project it gives too many linker errors.

How did you achieve to build the project ?

And which project file did you open exaxtly ?

Thanks a lot.

Berker.

--- (Edited on 1/15/2011 5:53 pm [GMT-0600] by enigma1987) ---

Re: Speech Recognition Model Converter
User: enigma1987
Date: 1/15/2011 5:59 pm
Views: 172
Rating: 15

And if possible could you please  send me the compiled version of the project in a tar file ?

my mail address is: [email protected]

maybe I could use the binary that you will send in my conversion work.

Thanks again.

Berker.

--- (Edited on 1/15/2011 5:59 pm [GMT-0600] by enigma1987) ---

Re: Speech Recognition Model Converter
User: dara
Date: 1/17/2011 11:44 am
Views: 130
Rating: 17

Hi enigma1987,


Yes, I got those exact errors before. It was because there're more than one function that have the same names. So, what I did was comenting the "cmd_ln_parse" and "cmd_ln_access" functions out in cmd_ln.c within "src/libs/s3_libutil/" directory. That means I use those two functions in the libcommon directory instead. Then, try to buid it again.

My executable file might not be suitable for you as I had edited some codes within the project to suit my need. For now, please try my suggestion above, it should work fine.

 

kind regards,

Dara

--- (Edited on 1/17/2011 11:44 am [GMT-0600] by dara) ---

Re: Speech Recognition Model Converter
User: enigma1987
Date: 1/17/2011 3:34 pm
Views: 147
Rating: 14

Hi Dara,

Thanks a lot.

I successfully compiled and got the exe file. 

I have three questions ?

1-) Will there be htkmmffn and htkmacrofn files (as empty) in the current directory before running the exe ? Or we will only give parameters and names as we like, and s3mod2htk will create the files ?

2-) I run the command and it gave an error:

Run-Time Check Failure #2 - Stack around the variable 'htkphone' was corrupted. Than I pressed the abort.

But it created hmmdefs and macros files. I dont know I did true or not ? I opened the two files and there are entries.

3-) What about tiedlist file ? How can I create it ?

I guess  I will need it in HTK and Julius when decoding.

Thanks again for your help.

Berker

--- (Edited on 1/17/2011 3:34 pm [GMT-0600] by enigma1987) ---

PreviousNext