Most, if not all biometric systems can be vulnerable to presentation attacks or spoofing. Spoofing involves the circumvention of a biometric system by a fraudster who masquerades as another enrolled client through the mimicking of their biometric traits. Spoofing attacks can be used to infiltrate systems or services protected by biometric technology, can undermine user confidence, and thus form a barrier to exploitation.
This talk will present recent work to develop presentation attack detection or countermeasure systems to protect speaker verification systems from diverse spoofing attacks including replay, speech synthesis and voice conversion. The talk will present the community-led ASVspoof initiative and the associated, competitive ASVspoof 2015 Challenge and results, alongside other common databases and evaluations, namely AVspoof and RedDots Replayed. The talk will cover progress in the field and present a treatment of constant Q cepstral coefficients (CQCCs) which are especially effective in distinguishing between genuine and spoofed speech. The talk will conclude with a brief discussion of current challenges and directions for future research.
