Automatic speaker verification (ASV) is among the most convenient, natural and efficient approaches to biometric authentication. Despite the appeal, and like the majority of all biometric systems, ASV is vulnerable to being manipulated by adversaries through spoofing or presentation attacks (PAs). Substantial interest in anti-spoofing or presentation attack detection (PAD) for ASV began only in 2013, with the first competitive challenge known as ASVspoof being held in 2015. Having attracted the participation of 62 independent teams for the 2019 edition, ASVspoof is now one of the largest community-supported anti-spoofing benchmarking challenges. This talk will describe the vulnerabilities of ASV systems to spoofing and PAs, the strategy behind ASVspoof as well as the progress made in the last five years. This shows that the best performing spoofing countermeasures or PAD systems can detect reliably even deepfake synthetic speech or converted voice attacks that humans are incapable of distinguishing from bona fide speech.
Nicholas Evans is a Professor in Audio Security and Privacy at EURECOM, a Graduate School and Research Centre in Digital Science located in
the French Riviera, the Mediterranean coast of southeastern France. His group's research interests include automatic speaker recognition, biometrics and both the security and privacy related implications including anti-spoofing/presentation attack detection, privacy preserving machine learning and anonymisation. He is among the founders of the Automatic Speaker Verification Spoofing and Countermeasures (ASVspoof) and VoicePrivacy community-led initiatives and challenges, is an associate editor of the IEEE Transactions on Biometrics, Behavior, and Identity Science and co-editor of the Handbook of Biometric Anti-Spoofing – Presentation Attack Detection.