Speech recognition relies on predefined grammars that determine valid speech input. Custom grammars allow you to limit the recognized words and phrases to those that make sense in the context of your app. Grammars allow you to improve the accuracy of speech recognition by limiting the set of recognizable phrases.
There are three types of grammars that can be used with the Windows Phone SDK:
Ready-to-use grammars, which include a dictation grammar and a web search grammar; both of which rely on a Microsoft cloud service
List grammars, which are lists of phrases that are defined programmatically by you
XML grammars, ...