I saw that somebody has already trained a german acoustic model for Sphinx! Great!
I was very curious and so I tested it using the HelloWorld Example of Sphinx4. I used small grammars containing digits and/or words like 'langsamer', 'schneller' etc that are listed in the vocabulary of the model. Unfortunately I have the problem, that the program always recognizes the same word regardless on what I say... If I change the grammar, another word is recognized again and again.
Has someone else tested it with sphinx4 and has an idea what could be wrong with my configuration?
For discussion of the model see sphinx forum:
There you can find a link on sphinx4 example:
The model is not perfect of course, so you are welcome to help.