Hi Mauricio,

>step by step instruction about how to create a dictionaries

In my opinion, the easiest program to use is for the creation of a pronunciation dictionary is: Sequitur G2P

What you do it take a pre-existing spanish dictionary (like h4.dict), and train "models" or "rules" that can create pronunications for unseen (also called out-of-vocabulary) words.  I used the following script:

#!/bin/sh

export PYTHONPATH=/usr/local/lib/python2.4/site-packages
time python ./g2p.py --train train.lex --devel 5% --write-model model-1
time python ./g2p.py --model model-1 --ramp-up --train train.lex --devel 5% --write-model model-2
time python ./g2p.py --model model-2 --ramp-up --train train.lex --devel 5% --write-model model-3
time python ./g2p.py --model model-3 --ramp-up --train train.lex --devel 5% --write-model model-4

time python ./g2p.py --model model-4 --ramp-up --train train.lex --devel 5% --write-model model-5

Change  PYTHONPATH to point to your Python packages.

Where train.lex is a pronunication dictionary (Spanish in your case) that looks something like this:

 A               ax
A'READY         ax r eh d iy
A'S             ey z
A.              ey
A.'S            ey z
A.S             ey z
A42128          ey f ao r t uw w ah n t uw ey t
AAA             t r ih p ax l ey
AABERG          aa b er g
AACHEN          aa k ax n
AAKER           aa k er

.... 

Then, after many hours, you can take a list of words that were not in you original pronunication dictionary, and create phonemes for them, as follows:

g2p.py --model model-5 --apply MissingWords 

Where MissingWords is a file containing a list of words you need pronunications for.

For instructions on how to create Monophone Acoustic Models, see Steps1-8 of the VoxForge Acoustic Model Tutorial. 

Creating a Triphone acoustic model (which provdies better recognition results) requires the creation of a tree.hed script (for HTK at least).   For more information on how to create a tree.hed file for a new language, see the following links:

• nsh's overview of how to create clustered triphone "questions" for Sphinx and HTK for new languages
• Ticket #153 - htk error on step 10, and a related thread in the forums
  
• my post on this Thread on "Error when compiling model" where I discuss Creating clustered triphone "questions"

See also the HTK manual. 

Ken 

Hi Ken:

>In my opinion, the easiest program to use is for the creation >of a pronunciation dictionary is: Sequitur G2P

>What you do it take a pre-existing spanish dictionary (like >h4.dict), and train "models" or "rules" that can create >pronunications for unseen (also called out-of-vocabulary) >words.  I used the following script:

>#!/bin/sh

>export >PYTHONPATH=/usr/local/lib/python2.4/site-packages
>time python ./g2p.py --train train.lex --devel 5% >--write-model model-1
>time python ./g2p.py --model model-1 --ramp-up >--train train.lex --devel 5% --write-model model-2
>time python ./g2p.py --model model-2 --ramp-up >--train train.lex --devel 5% --write-model model-3
>time python ./g2p.py --model model-3 --ramp-up --train train.lex --devel 5% --write-model model-4

>time python ./g2p.py --model model-4 --ramp-up >--train train.lex --devel 5% --write-model model-5

>Change  PYTHONPATH to point to your Python packages.

>Where train.lex is a pronunication dictionary (Spanish in your >case) that looks something like this:

 ok, so I can take the h4.dict as the train.lex file?

>Then, after many hours, you can take a list of words that >were not in you original pronunication dictionary, and create >phonemes for them, as follows:

>g2p.py --model model-5 --apply MissingWords 

>Where MissingWords is a file containing a list of words you >need pronunications for.

ok, so the "MissingWord" is the output file of the above procees with the train.lex (h4.dict)?

Is this kind of work needed for spanish support?, or kind of ready with the work in:

http://www.speech.cs.cmu.edu/sphinx/models/hub4spanish_itesm/

or need more words to be added, etc? the idea is to do missing works  or things that are missing...

thanks for your help..

bye

Mauricio

Hi Mauricio,

> ok, so I can take the h4.dict as the train.lex file?

Yes 

>ok, so the "MissingWord" is the output file of the above procees with the

>train.lex (h4.dict)?

No, the MissingWords file is simply a list of words that do *not* exist in your pronunciation dictionary.   G2P reads each word in the MissingWords file (one word per line) and creates/generates a pronunciation for it.

>Is this kind of work needed for spanish support?, 

It depends ... if the words in your current pronunciation dictionary meet all your grammar needs, then you don't need to add any new words. 

But, if you want to create a new grammar for a specific application that uses words that are not in your pronunciation dictionary, then you need to add the pronunciations either manually or use Sequitor G2P to automate the process.

Where I use this quite a bit is when I want to process audiobooks from the LibriVox project.  The text that the reader has recorded always has many words that are not included in the VoxForge pronunciation dictionary. In order to process this audio to include them in the acoustic model creation process, I need to have pronunciations for these words.  Sequitor G2P saves a lot of time.

Ken