Hey Guys... 

I have trained a Spanish model that works pretty well. The model is for decoding 8kHz 16Bits recordings. 

Decoding stage results were:

Aligning results to find error rate

        SENTENCE ERROR: 14.2% (99/699)   WORD ERROR RATE: 3.0% (316/10688)

Inside the folder etc you can find the file sphinxtrain.cfg that  I used.

Enjoy...

Akane.

VOXFORGE_ES_8KHZ.zip