Practical lessons of (deep)faking human speech

deepfakes, biometrics, spoofing attack, impersonation

Deepfakes are an emerging threat to computer security. Various usages range from spreading fake news to identity theft. A particular subset of this problem is using deepfakes to spoof biometrics systems. The attacker spoofs an individual's identity by synthesizing his voice or appearance and uses this spoofed identity to gain access into a system secured by biometrics authentication. In our former research, we demonstrated that speech deepfakes present a threat not only to voice biometrics systems but also to people. We presented that text-dependent verification is more resilient to deepfakes than text-independent verification. This paper aims to summarize our former research with its results and to highlight the powerful role of open-source tools in this area.