Speech Processing in Realistic Battlefield Environments (Le Traitement de la Parole en Environnement de Combat Realiste)

摘要

Communications, command and control, intelligence, and training systems are more and more making use of speech technology components: i.e., speech coders, voice controlled C2 systems, speaker and language recognition, and automated training suites. Interoperability of these systems is not a simple standardization problem as the speech of each individual user is an uncontrolled variable such as non-native speakers using, in addition to their own language, an official NATO language. For multi-national military operations, this may cause a reduced performance or even cause malfunction of an action. Standardized assessment methods and specifications both for commercial-off-the- shelf (COTS) and for development of new technology are required. The work was separated into four tasks: (1) Collect native and non-native unclassified and classified speech communications from training exercises and actual operations; (2) Produce annotated database(s) that might be used beyond the confines of the Task Group; (3) Assess effects on performance of recognisers and communication equipment; and (4) Relate derived results to military applications. In this report the results of the study are presented.