In order to be able to develop and test techniques in improving spoken dialog systems, a baseline system was built. CMU has significant experience in building spoken dialog systems. Since we want to have significant control of the dialog system, completely off-the-shelf systems supporting VoiceXML-like directed dialog would be too restrictive. We therefore chose to build on the RavenClaw [3] system. This system is in turn built upon the MIT Galaxy architecture [4] and uses the CMU Sphinx speech recognizer [5] and the Festival Speech Synthesis System [6].