Audio-deepfake detection: Adversarial attacks and countermeasures

<p>Audio has always been a powerful resource for biometric authentication: thus, numerous AI-based audio authentication systems (classifiers) have been proposed. While these classifiers are effective in identifying legitimate human-generated input their security, to the best of our knowledge,...

وصف كامل

محفوظ في:
التفاصيل البيبلوغرافية
المؤلف الرئيسي: Mouna Rabhi (17086969) (author)
مؤلفون آخرون: Spiridon Bakiras (16896408) (author), Roberto Di Pietro (16864155) (author)
منشور في: 2024
الموضوعات:
الوسوم: إضافة وسم
لا توجد وسوم, كن أول من يضع وسما على هذه التسجيلة!
_version_ 1864513516016238592
author Mouna Rabhi (17086969)
author2 Spiridon Bakiras (16896408)
Roberto Di Pietro (16864155)
author2_role author
author
author_facet Mouna Rabhi (17086969)
Spiridon Bakiras (16896408)
Roberto Di Pietro (16864155)
author_role author
dc.creator.none.fl_str_mv Mouna Rabhi (17086969)
Spiridon Bakiras (16896408)
Roberto Di Pietro (16864155)
dc.date.none.fl_str_mv 2024-09-15T03:00:00Z
dc.identifier.none.fl_str_mv 10.1016/j.eswa.2024.123941
dc.relation.none.fl_str_mv https://figshare.com/articles/journal_contribution/Audio-deepfake_detection_Adversarial_attacks_and_countermeasures/25827097
dc.rights.none.fl_str_mv CC BY 4.0
info:eu-repo/semantics/openAccess
dc.subject.none.fl_str_mv Information and computing sciences
Artificial intelligence
Cybersecurity and privacy
Authentication
Adversarial attacks
Audio deepfake
Fake voice detection
GAN
Biometrics
Security
dc.title.none.fl_str_mv Audio-deepfake detection: Adversarial attacks and countermeasures
dc.type.none.fl_str_mv Text
Journal contribution
info:eu-repo/semantics/publishedVersion
text
contribution to journal
description <p>Audio has always been a powerful resource for biometric authentication: thus, numerous AI-based audio authentication systems (classifiers) have been proposed. While these classifiers are effective in identifying legitimate human-generated input their security, to the best of our knowledge, has not been explored thoroughly when confronted with advanced attacks that leverage AI-generated deepfake audio. This issue presents a serious concern regarding the security of these classifiers because, e.g., samples generated using adversarial attacks might fool such classifiers, resulting in incorrect classification. In this study, we prove the point: we demonstrate that state-of-the-art audio deepfake classifiers are vulnerable to adversarial attacks. In particular, we design two adversarial attacks on a state-of-the-art audio-deepfake classifier, i.e., the Deep4SNet classification model, which achieves 98.5% accuracy in detecting fake audio samples. The designed adversarial attacks 1 1 The code of the attacks will be released open-source in the camera ready. leverage a generative adversarial network architecture and reduce the detector’s accuracy to nearly 0%. In particular, under graybox attack scenarios, we demonstrate that when starting from random noise, we can reduce the accuracy of the state-of-the-art detector from 98.5% to only 0.08%. To mitigate the effect of adversarial attacks on audio-deepfake detectors, we propose a highly generalizable, lightweight, simple, and effective add-on defense mechanism that can be implemented in any audio-deepfake detector. Finally, we discuss promising research directions.</p><h2>Other Information</h2> <p> Published in: Expert Systems with Applications<br> License: <a href="http://creativecommons.org/licenses/by/4.0/" target="_blank">http://creativecommons.org/licenses/by/4.0/</a><br>See article on publisher's website: <a href="https://dx.doi.org/10.1016/j.eswa.2024.123941" target="_blank">https://dx.doi.org/10.1016/j.eswa.2024.123941</a></p>
eu_rights_str_mv openAccess
id Manara2_134d8e61a1bd07f48f44a9756a52e866
identifier_str_mv 10.1016/j.eswa.2024.123941
network_acronym_str Manara2
network_name_str Manara2
oai_identifier_str oai:figshare.com:article/25827097
publishDate 2024
repository.mail.fl_str_mv
repository.name.fl_str_mv
repository_id_str
rights_invalid_str_mv CC BY 4.0
spelling Audio-deepfake detection: Adversarial attacks and countermeasuresMouna Rabhi (17086969)Spiridon Bakiras (16896408)Roberto Di Pietro (16864155)Information and computing sciencesArtificial intelligenceCybersecurity and privacyAuthenticationAdversarial attacksAudio deepfakeFake voice detectionGANBiometricsSecurity<p>Audio has always been a powerful resource for biometric authentication: thus, numerous AI-based audio authentication systems (classifiers) have been proposed. While these classifiers are effective in identifying legitimate human-generated input their security, to the best of our knowledge, has not been explored thoroughly when confronted with advanced attacks that leverage AI-generated deepfake audio. This issue presents a serious concern regarding the security of these classifiers because, e.g., samples generated using adversarial attacks might fool such classifiers, resulting in incorrect classification. In this study, we prove the point: we demonstrate that state-of-the-art audio deepfake classifiers are vulnerable to adversarial attacks. In particular, we design two adversarial attacks on a state-of-the-art audio-deepfake classifier, i.e., the Deep4SNet classification model, which achieves 98.5% accuracy in detecting fake audio samples. The designed adversarial attacks 1 1 The code of the attacks will be released open-source in the camera ready. leverage a generative adversarial network architecture and reduce the detector’s accuracy to nearly 0%. In particular, under graybox attack scenarios, we demonstrate that when starting from random noise, we can reduce the accuracy of the state-of-the-art detector from 98.5% to only 0.08%. To mitigate the effect of adversarial attacks on audio-deepfake detectors, we propose a highly generalizable, lightweight, simple, and effective add-on defense mechanism that can be implemented in any audio-deepfake detector. Finally, we discuss promising research directions.</p><h2>Other Information</h2> <p> Published in: Expert Systems with Applications<br> License: <a href="http://creativecommons.org/licenses/by/4.0/" target="_blank">http://creativecommons.org/licenses/by/4.0/</a><br>See article on publisher's website: <a href="https://dx.doi.org/10.1016/j.eswa.2024.123941" target="_blank">https://dx.doi.org/10.1016/j.eswa.2024.123941</a></p>2024-09-15T03:00:00ZTextJournal contributioninfo:eu-repo/semantics/publishedVersiontextcontribution to journal10.1016/j.eswa.2024.123941https://figshare.com/articles/journal_contribution/Audio-deepfake_detection_Adversarial_attacks_and_countermeasures/25827097CC BY 4.0info:eu-repo/semantics/openAccessoai:figshare.com:article/258270972024-09-15T03:00:00Z
spellingShingle Audio-deepfake detection: Adversarial attacks and countermeasures
Mouna Rabhi (17086969)
Information and computing sciences
Artificial intelligence
Cybersecurity and privacy
Authentication
Adversarial attacks
Audio deepfake
Fake voice detection
GAN
Biometrics
Security
status_str publishedVersion
title Audio-deepfake detection: Adversarial attacks and countermeasures
title_full Audio-deepfake detection: Adversarial attacks and countermeasures
title_fullStr Audio-deepfake detection: Adversarial attacks and countermeasures
title_full_unstemmed Audio-deepfake detection: Adversarial attacks and countermeasures
title_short Audio-deepfake detection: Adversarial attacks and countermeasures
title_sort Audio-deepfake detection: Adversarial attacks and countermeasures
topic Information and computing sciences
Artificial intelligence
Cybersecurity and privacy
Authentication
Adversarial attacks
Audio deepfake
Fake voice detection
GAN
Biometrics
Security