An assessment of automatic speaker verification vulnerabilities to replay spoofing attacks
Artur Janicki , Federico Alegre , Nicholas Evans
AbstractThis paper analyses the threat of replay spooﬁng or presentation attacks in the context of automatic speaker veriﬁcation. Asrelatively high-technology attacks, speech synthesis and voice conversion, which have thus far received far greater attentionin the literature, are probably beyond the means of the average fraudster. The implementation of replay attacks, in contrast,requires no speciﬁc expertise nor sophisticated equipment. Replay attacks are thus likely to be the most proliﬁc in practice,while their impact is relatively under-researched. The work presented here aims to compare at a high level the threat ofreplay attacks with those of speech synthesis and voice conversion. The comparison is performed using strictly controlledprotocols and with six different automatic speaker veriﬁcation systems including a state-of-the-art iVector/probabilisticlinear discriminant analysis system. Experiments show that low-effort replay attacks present at least a comparable threatto speech synthesis and voice conversion. The paper also describes and assesses two replay attack countermeasures. Arelatively new approach based on the local binary pattern analysis of speech spectrograms is shown to outperform acompeting approach based on the detection of far-ﬁeld recordings.
|Journal series||Security and Communication Networks, ISSN 1939-0114 [1939-0122]|
|Publication size in sheets||0.7|
|Keywords in English||speaker veriﬁcation; spooﬁng; presentation attack; replay; countermeasures; local binary patterns|
|project||The Develpment of Digital Communicatios. Project leader: Siuzdak Jerzy,
, Phone: +48 22 234-7232, start date 27-04-2015, end date 31-12-2016, IT/2015/statut, Completed
|Score|| = 20.0, 27-03-2017, ArticleFromJournal|
= 20.0, 27-03-2017, ArticleFromJournal
|Publication indicators||: 2016 = 1.067 (2) - 2016=1.022 (5)|
|Citation count*||5 (2018-02-22)|
* presented citation count is obtained through Internet information analysis and it is close to the number calculated by the Publish or Perish system.