In accordance with the present invention, using client-server communications over a packet network, such as the Internet, the ASR server receives a grammar from the client, receives information representing speech from the client, performs speech recognition, and the TTS server returns information based upon the recognized speech to the client.