A system and method of operating an automatic speech recognition
application over an Internet Protocol network is disclosed. The ASR
application communicates over a packet network such as an Internet
Protocol network or a wireless network. A grammar for recognizing
received speech from a user over the IP network is selected from a
plurality of grammars according to a user-selected application. A server
receives information representing speech over the IP network, performs
speech recognition using the selected grammar, and returns information
based upon the recognized speech. Sub-grammars may be included within the
grammar to recognize speech from sub-portions of a dialog with the user.