Acceptance power of Speech Recognition Grammars

Hi,

I was reading specifications for Speech Recognition Grammars, it seems to me
these are Regular Grammars. If this is true, it will accept only subset of
Natural Languages.

Here is the list of Grammars with the acceptance power low to high.

Regular Grammars : Accepts Regular Languages (Pattern Recognition)
Context-Free Grammars : Accepts Context-Free Languages (Computer Languages)
Context-Sensitive Grammars : Accepts Context-Sensitive Languages (Natural
Languages)
Turing Machines : Accepts Recursive and Recursively Enumerable Sets
(Computation)

Is it a limitation on the Speech Recognition Grammars ?

Thanks,
Sudhakar.