Reconstructing Human Physiological Features from Audio Samples
Abstract
The unique sound of our voices is a direct byproduct of the sizes and shapes of ouranatomical features. Our oral cavity, chest, nose, and even lips influence how we sound. As a result, our physical features impact our unique voice fingerprints. The literature is rich with work correlating components of the vocal tract with to their outputs. However, none have looked at the reverse case, where voices serve as a means of reconstructing physical features. This project builds facial models of speakers, both known and unknown, using techniques from fluid dynamics. Such techniques have potentially high impact not only in the space of authentication, but also in detecting audio deep fakes.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 05, 2021
- Source ID
- N000142112658
Entities
People
- Patrick Traynor
Organizations
- Office of Naval Research
- United States Navy
- University of Florida