The Design and Requirements Evolution of a Speech Recognition Technology for Tactical Applications and Environments
Abstract
This paper presents a discussion of the unique and specialized requirements for a militarized speech recognizer. It also presents the tactical advantages of militarized speech recognition technology, as it could be applied in several Command and Control (C2) applications and environments. Additionally, this paper will present the results of a comparison study, which was performed between a custom military speech recognition technology and various manual input modalities, including keyboard and trackball, for activating a selected C2 application. The results of this paper demonstrate a clear superiority of continuous speech recognition over discrete speech recognition in both metrics, and a tradeoff of task execution speed for error rate for continuous speech recognition verses manual input normal operation was evaluated at normal environmental levels. Helicopter environments are relatively stationary. The environment does not spectrally vary enough during various flight profiles to affect recognition performance. However, the noise environment of military track vehicles is not nearly as stationary as the helicopter environment, and these variations must be taken into consideration. For our own work we wanted to eliminate the user from having to enroll in the environment altogether, and we wanted to eliminate environmental dependency. Our approach to eliminating the noise during the enrollment session (at least as far as the test subject was concerned) was to electrically mix environmental noise into the test subject input stream to the target speech recognizer. The RADC work indicated that the precise signal-to-noise ratio should not be too critical, and indeed, through
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 2004
- Accession Number
- ADA432741
Entities
People
- Lockwood Reed
Organizations
- United States Army