Voice Biometrics for Information Assurance Applications
Abstract
In 2002, the President of the United States established an organization within the DOD to develop and promulgate biometrics technologies to achieve security in information, information systems, weapons, and facilities. NRL has been tasked to study voice biometrics for applications in which other biometrics techniques are difficult to apply. The ultimate goal of voice biometrics is to enable the use of voice as a password. Voice biometrics are "man-in-the-loop" systems in which system performance is significantly dependent on human performance. This aspect has not been properly emphasized by previous researchers in this field. Accordingly, we let each speaker choose his (or her) own test phrase that can be uttered consistently. The speech waveform is then pre-processed (i.e, equalized and normalized) to reduce the effect of inconsistent speaking. Subsequently, we extract five different voice features from the speech waveform. Some of them have never been used for voice biometrics. Finally, individual feature errors are combined to indicate a confidence level of speaker verification. Initial laboratory testing under various conditions shows encouraging results. We will be prepared to fleet-test our voice biometrics system in FY03.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 05, 2002
- Accession Number
- ADA408449
Entities
People
- George S. Kang
- Yvette Lee
Organizations
- United States Naval Research Laboratory