Evaluation and Improvement of a Speaker Verification System in Military Environments
Abstract
The U.S. Navy is constantly looking for ways to improve security. One of the possible solutions proposed has been the use of biometrics, defined by the Biometric Consortium as "automated methods of recognizing a person base on a physiological or behavioral characteristic". One of the least costly, most convenient, and least invasive methods of biometrics is speaker verification. The Naval Research Laboratory's Voice Systems Section has undergone an extensive study of a commercial speaker verification system in adverse military noise and voice encoding environments. Additional testing was conducted to study the amount of improvement that could be achieved using a noise canceling preprocessor. This report documents the performance results in 10 different military noise environments, six different voice encoding algorithms, and all combinations of the two, with and without noise cancellation. Results show that speaker verification definitely show promise under some conditions. In many cases, a significant improvement in performance was achieved by using a noise cancellation preprocessor. While more testing needs to be done to gauge the level of performance under live conditions, this report shows system integrators where speaker verification could possibly be used.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 26, 2002
- Accession Number
- ADA409777
Entities
People
- David A. Heide
Organizations
- United States Naval Research Laboratory