One of the limitations of most commercial speech-to-text systems is that they need to be trained. They also are limited in that they usually work with only the person who trained (one participant). This is suitable for applications like broadcast news transcripts, however, in meetings there are many people and there is no training possible. In AMI, several partners are working on Large Vocabulary Conversational Speech Recognition (LVCSR) systems. In this demonstration the participant can type in a word that should be recognized and then begin recording a multiparty meeting. The LVCSR will recognize the word in real time and alert the system user. This is particularly compelling when in a meeting because approximately 40% of the verbal communication between people is overlapping. This means that the LVCSR has to detect a word and attribute it correctly to the person who spoke it.