Document Type


Publication Date





This paper describes the development of a real-time automated transcription tool for assessing tactical communications in a DIS environment. Java-based tools were developed to capture simulated radio communications data from tactical training exercises conducted at the Warfighter Training Research Division of the Air Force Research Laboratory’s Human Effectiveness Directorate. A representative set of audio data was handtranscribed and used as training material for a class-based statistical language model using a commercially available speech recognition system. The language model was designed to allow run-time input of callsign data to provide increased flexibility. The resulting system is a real-time automated speech-to-text transcription tool that logs the audio data obtained from signal PDUs as a standard wave file and produces a text transcription, aiding in assessing tactical communications effectiveness. To test the capabilities of this system, an evaluation was performed using DIS log data from similar training exercises. Preliminary results indicate that overall word error rates across all participants were around 18%. For individual stations, however, word error rates less than 4% were obtained. Additional efforts are underway to refine the language model to achieve further reductions in error rate. Also discussed are efforts to further develop the Java DIS tools to provide scanner, logger, and basic radio functionality.