Acoustic model based on letters rather than words

[puneet]

hi
I am trying to build acoustic model based on phones.
i have prepared a dictionary of letters in which all their corresponding phones are written.
For eg.. for the word Hello.... In dictionary I am wiriting.....
He He
llo llo
I am doing in my native language not english.
this is being done so that i can derive words from the combination of different letters.
my problem is how to build a language model and combine the words.
Please help

[Nickolay V. Shmyrev]

There is no such thing as "model based on phones" since acoustic model describes sounds of the language, not phones or words.

You need to follow standard model training tutorial from here

http://cmusphinx.sourceforge.net/wiki/tutorialam

If you have questions feel free to ask. Dictionary creation is covered here

http://cmusphinx.sourceforge.net/wiki/tutorialdict

[puneet]

Actually I want to develop model based on sub word units
So that out of vocabulary words can be recognized
I have build dictionary and phone file
In dictionary i have trained all phones (i.e.sub words)
now i want to combine phones to get the full word.

Any thing possible in sphinx regarding this approach.

[Nickolay V. Shmyrev]

Phonetic recognition is possible, described in our wiki. You can also use syllable-based language models.

[puneet]

thanks a lot

[puneet]

Hi Nickolay
Kindly send me some link where I can read about hoe to build syllable-based language models.
Thanks...

[Nickolay V. Shmyrev]

Take a text, convert every word to syllables with the dictionary, for example (HELLO -> HH_EH L_OW), then train a language model as usual.