Automated gisting systems for voice communications

Abstract

It is pointed out that the detection of a limited number of key words in voice communications, combined with nonlinguistic cues and situation knowledge, holds out the promise for automatic extraction of the general content or gist of the transmitted messages. A systems-level description of an end-to-end automatic gisting system for screening voice communications is presented. The emphasis is on identifying information which can usefully supplement the detected key words in the gisting process. Principal subsystems are identified, and their functions are discussed in the context of an overall system architecture. The example of automatic determination of aircraft takeoffs and landings at an airport, based on monitoring conversations between pilots and air traffic controllers, is used for illustrative purposes.<>