An important part of meeting recognition is the ability to efficiently capture, manipulate and review all aspects of a meeting. To that end we have developed a meeting browser that lets users:
Create meeting records and transcriptions of meetings with participants remotely located.
Create and customize dialogue, audio, and video summaries to the user's particular needs.
Create a database of corporate knowledge.
Quickly and accurately create and disseminate a list of conclusions and action items
Provide rapid access to meeting records to allow browsing and reviewing existing meetings.
Identify for each utterance the speaker properties (type, social relationships, and emotion) as well as the discourse structure and type.
When a meeting is being created, each participant may join either remotely or locally. Once the meeting has begun, speech is transmitted to Janus, our speech recognition engine. As the speech is recognized, the hypothesis is sent to the dialogue system where it is assembled into a meeting format. The meeting browser displays the transcript for the current meeting. The meeting transcript can be sent to the summarization system which will create a summary of the current dialogue. Finally, a user may elect to save a meeting including any summaries in the meeting archive from within the meeting browser.

At the end of meetings, it is customary to reiterate a set of action items. Using speech recognition, we recognize the items and mail them out to each of the meeting participants. Likewise, we can mail complete meetings, meeting segments, or summaries including the audio portion directly from within the meeting browser to meeting participants or any other interested parties. Each of these may include annotations, comments or corrections. Corrections can be done by using a keyboard or handwriting recognition using a handwriting recognizer developed in our lab. In the future we plan to add speech recognition as an additional error repair modality.