Speech recognition converts spoken words to string, and I do not think it is possible, nor suitable, to entangle with the speech recognition engine for that. Moreover, you cannot recognize anything before having it parsed in the first place (from audio signal to actual string).
Real-One wrote:
I think it will be better to constrain the spoken words to only numbers before recognizing and not after.
You may have to reconsider this opinion :)
Kindly.