The vulnerability of automatic speaker verification (ASV) systems to spoofing attacks is now well acknowledged. If exposed, these vulnerabilities not only threaten reliability, but also have potential to damage user-confidence and hence hinder the future exploitation of ASV technology. It is thus critical that countermeasures to protect from spoofing are integrated into the OCTAVE platform or trusted biometric authentication service (TBAS). The first set of OCTAVE spoofing countermeasures were reported in the companion deliverable, D9 “Robustness to spoofing with existing countermeasures”. Results reported therein demonstrate the potential to detect and deflect certain spoofing attacks with reasonable reliability. Since spoofing attacks in the wild are likely to evolve over time, however, and since spoofers will inevitably change tactics in response to deployed countermeasures, it is essential that the development of countermeasures keep apace with the evolving sophistication of spoofing attacks. This deliverable is therefore a follow-up to D9 and reports the OCTAVE consortium’s progress in anti-spoofing. Reported is a second set of advanced countermeasures and related studies that either deliver enhanced detection performance, better generalization to unforeseen attacks, or develop our understanding of likely, practical performance in OCTAVE’s user-case scenarios. Enhanced detection performance is required to improve reliability to even the most sophisticated spoofing algorithms. Generalisation will help protect today’s systems from tomorrow’s attacks. Other, advanced studies reported here fill gaps in the current technology and understanding. The literature lacks sufficient attention to replay attacks that will likely be the most prolific form of attack in the wild. The performance of existing countermeasures in the face of varying acoustic and system conditions also needs to be assessed. The current deliverable thus presents pioneering new studies in addition to the latest state-of-the-art developments in spoofing countermeasures. Among these is a number of new countermeasures that deliver significant improvements to spoofing detection across three common databases, several new approaches to replay detection and the first studies of the impact of bandwidth and additive noise on detection performance. Findings from all of this work will be used to inform the selection of countermeasure candidates and their configuration for integration into the OCTAVE TBAS later in the project.
Source: WP 3 Robustness in Speaker Verification
Dissemination level: Confidential
To know more about the document, you may place a request in the ‘Contact’ section of this site. We reserve the right to decide how much we can disclose.