The language model you were using is not for broadcast news, it's a small test language model to test the acoustic model quality. It is not supposed to recognize generic type of speech.

I thought you were going to use broadcast news language model, it is more reasonable.

Alternatively, you can try to create your own language model from wikipedia texts.

Sorry, that is the wrong model. I had to revert to trying to use it instead because the broadcast news model is for Mexican Spanish, and as you had pointed out, Mexican Spanish couldn't be used with the Voxforge model.

I think I would actually like to create a model of my own. Do you know if there are tools to create one with the Wikipedia texts?

> I had to revert to trying to use it instead because the broadcast news model is for Mexican Spanish, and as you had pointed out, Mexican Spanish couldn't be used with the Voxforge model.

You misunderstood me. Please read the tutorial about the concepts of speech recognition to learn the difference between the acoustic model, the dictionary and the language model:

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

I only told you to use the Spanish dictionary, you still can use broadcast news language model with the Spanish Spanish dictionary instead of Mexican Spanish dictionary.

> Do you know if there are tools to create one with the Wikipedia texts?

Please read

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

http://trulymadlywordly.blogspot.ru/2011/03/creating-text-corpus-from-wikipedia.html