Meeting Recognition and Understanding

Établi : July 30, 2013

In most organizations, staff spend many hours in meetings. This project addresses all levels of analysis and understanding, from speaker tracking and robust speech transcription to meaning extraction and summarization, with the goal of increasing productivity both during the meeting and after, for both participants and nonparticipants.

The Meeting Recognition and Understanding project is a collection of online and offline spoken language understanding tasks. The following functions could be performed both on- and offline, but generally work best retrospectively, with the full meeting available. The overall integration of these technologies result in a structured, searchable, and cross-referenced document that can be integrated in corporate unified communications infrastructure.

Automatic speech recognition
Speaker diarization («who spoke when?»)
Addressee detection («who spoke to whom?»)
Sentence segmentation and disfluency cleanup
Dialog act tagging
Named entity extraction
Extracting distinguishing keyphrases
Topic segmentation
Topic identification (optionally with agenda)
Keyword spotting
Hot spot detection
Speaker role detection
Argument diagramming for meeting structure extraction
Agreement/disagreement detection
Extraction of action items and decisions
Meeting summarization

Other technologies allow meeting participants to interact with a virtual meeting participant for online assistance. These allow participants to (explicitly or implicitly) call up information that is relevant to the meeting, or flag content on-the-fly (such as action items and decisions).

Situational spoken language understanding (includes semantic parsing, intent determination, and click/object detection)
Dialog manager
Addressee detection

Our work builds on ongoing work in Conversational Understanding research, and past work on the ICSI Meeting Recorder (opens in new tab) and DARPA The CALO Meeting Assistant System projects.