The need for a robust digitized database to support tactical speech processing, research & development

Describes a new database developed for Rome Laboratory, called the GISTING Speech Processing Database, which solves many of the problems associated with previous tactical speech databases. Unlike some previous speech databases, this database provides for statistically significant samples to perform system training. Similarly, sufficient data exist to ensure that scientists involved in the development of keyword spotting algorithms will be able to experiment with a wide variety of real-world internal and external factors. In all, the database contains over 100 fields reflecting internals, externals, statistics, measurements, and analytical comments related to a selected sound section. It is concluded that experiments using this database could suggest new or modified technical approaches which will provide the breakthroughs needed to optimize or develop speech processing algorithms which can be effectively used in a tactical environment