The telephony SR engines used by Speech Server do not self-train.
If you are asking about the desktop recognition engines, please go to
http://www.microsoft.com/communities/newsgroups/en-us/, locate the "Speech Desktop" newsgroup in the tree at English > Office and Desktop Applications > Speech Desktop, and repost your question there.
These postings are provided "AS IS" with no warranties, and confer no rights.