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 and DARPA The CALO Meeting Assistant System projects.