I-Corps: Liveness detection and integrity authentication of digital audio

Regents Of The University Of Michigan - Dearborn, Dearborn MI

Investigators

Abstract

The broader impact/commercial potential of this I-Corps project is the development of a reliable liveness detection and integrity authentication tool for digital audio. Recent technological advances in the areas of artificial intelligence, fake audio generation, and easy access to smart speakers pose serious security and privacy threats to voice activated devices/services (VAD/S)-based systems. Researchers have demonstrated existing VAD/S are vulnerable to attacks and the growing adoption of VAD/S for e-commerce, voice-based search, and speaker recognition-based banking and remote identity verification highlights the problem. The proposed technology aims to secure VAD/S, e.g., Google Home, Amazon Alexa, Apple Siri, etc., against a growing number of attacks including spoofing, replay, and deepfake attacks. Specifically, the proposed technology may reliably detect and prevent attacks on speaker/speech recognition systems using liveness detection for real-time systems and determine integrity verification for off-line systems. In addition, the proposed technology may benefit other areas including multimedia forensics, e-discovery, remote biometric verification, law enforcement and the entertainment industry. This I-Corps project is based on the development of real-time liveness detection and integrity authentication of digital audio data. This is a relevant problem that concerns security of many critical applications central to modern digital usage such as speaker verification, speech recognition, fintech, e-commerce, and social medial platforms. Reliable liveness detection may be used to counter the threat of deepfakes to speaker verification and the spread of disinformation and falsehood on social media platforms while safeguarding speaker verification systems commonly used for noninvasive identity verification. The proposed liveness detection and integrity verification of audio data framework relies on the physics of acoustic and photo-acoustic processes and mathematical modeling of distortion artifacts due replay attacks. The solution will use these to detect liveness of audio with the input of voice activated devices and services. In addition, the proposed innovation leverages the acoustic environment, acquisition device, and post-processing artifacts for forensic analysis and content integrity authentication. This technology may protect VAD/S against the growing number of attacks and provide forensic examiners and law enforcement personnel with a powerful, computationally efficient, scalable, and reliable tool to perform online as well as offline forensics analysis. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.