After reading through this site, I see that this project seems only to be aimed at creating a solid acoustic model for various software. I'm working with Juilis, and was looking into the grammar files wondering how to expand on them easily.

I also see that this site has collected a fair amount of speech and associated text. Is it possible, and if so why is it not being done: Grammer files generated from the input text and audio.

I realize this may not be a clear question, but I think I get the general idea across.

--- (Edited on 4/11/2010 8:47 pm [GMT-0500] by ) ---

>is it possible, and if so why is it not being done: Grammer files generated

>from the input text and audio.

From step 1 of the VoxForge Tutorial

Language Model or Grammar - [Statistical] Language Models contain a very large list of words and their probability of occurrence in a given sequence.  They are used in dictation applications.  Grammars are a much smaller file containing sets of predefined combinations of words.  Grammars are used in IVR or desktop Command and Control applications.   Each word in a Language Model or Grammar has an associated list of phonemes (which correspond to the distinct sounds that make up a word).

The text used for the prompts in the VoxForge speech submission applet is focused on collecting phonemes, and does not reflect speech (word sequences, phrases...) from a particular domain...

For example, if you want to recognize speech from the medical domain, you create a statistical language model (in addition to creating an acoustic model) from words and phrases from that 'domain', like papers, emails, transcribed speeches...

Ken

--- (Edited on 4/11/2010 10:57 pm [GMT-0400] by kmaclean) ---

Would it not make sense however, to at least leverage this great community input to help build at least a general language model/grammar as a baseline for people who want to utilize this contribution?

Heck, even if it were possible to even be able to select words and get the phonemes for them, instead of having to do all that work by hand, that would be so much more useful. (Unless I've missed something and there is a way to automatically generate them from input speech/text)

--- (Edited on 4/12/2010 5:50 am [GMT-0500] by ) ---

>if it were possible to even be able to select words and get the phonemes

>for them

I am still not sure I understand what you are asking... Do you mean like this?:

http://www.speech.cs.cmu.edu/cgi-bin/cmudict

--- (Edited on 4/12/2010 9:50 am [GMT-0400] by kmaclean) ---

It would seem so yes. Maybe a good idea would be to link this resource somewhere in the about section?

--- (Edited on 4/12/2010 8:55 am [GMT-0500] by ) ---

>Maybe a good idea would be to link this resource somewhere in the about

>section?

The CMU site does not use the same phone list as VoxForge phone list.

You can get the list of pronunciations from the lexicon file in Step 2 of the Tutorial, but that only contains a small number of words (around 27000 words). 

For a pronunciation dictionary with many more words, use  http://www.dev.voxforge.org/projects/SpeechCorpus/browser/Trunk/Lexicon/VoxForge/VoxForgeDict (131395 words), which is what the VoxForge acoustic models are trained on.

But you will need to recreate your acoustic models using this pronunciation dictionary (and update the pronunciations in your grammar file).

--- (Edited on 4/12/2010 12:23 pm [GMT-0400] by kmaclean) ---

Thanks for the help.

One more related question. Is there some sort of tool or program that could be used to dynamically add words that were not recognized, and update all relevant files and models? Aka, a way to retrain software on the fly?

--- (Edited on 4/12/2010 11:36 am [GMT-0500] by bendauphinee) ---

Also, I see you linked the Linux tutorial. How would I incorporate that data in a Windows environment, since I don't currently have access to a Linux rig?

--- (Edited on 4/12/2010 11:46 am [GMT-0500] by bendauphinee) ---

>Is there some sort of tool or program that could be used to dynamically

>add words that were not recognized,

see Simon - it uses Julius

--- (Edited on 4/12/2010 2:09 pm [GMT-0400] by kmaclean) ---