Sniffing out the fakes – Google files patent apps for fingerprint and voice liveness
Google has filed a pair of patent applications for software capable of detecting spoofed voice and fingerprint biometrics that could be used to gain access to facilities and information.
The applications are 20230385393 A1, relating to localized illumination for biometric authentication, and 20230386506 A1, for self-supervised speech representations for detecting fake audio. Both were published November 30.
The first one involves a fingerprint scanner that would change the luminosity and color of multiple regions of the screen.
Biometric authentication managing software would create illumination patterns to create and analyze a print reflection. The reflection would be used to detect print authenticity and liveness through characteristics like blood oxygen saturation or heart rate.
The second patent application involves plans for software and a device capable of recording audio data to spot synthetic speech.
In engineers’ plans, involves using a trained self-supervised model to generate “audio feature vectors.” The data would be scored by a shallow discriminator model looking for signs of a synthetic voice.
The score would be based on the features of each audio feature vector.