Protecting gender and identity with disentangled speech representations

被引:6
|
作者
Stoidis, Dimitrios [1 ]
Cavallaro, Andrea [1 ]
机构
[1] Queen Mary Univ London, Ctr Intelligent Sensing, London, England
来源
基金
英国工程与自然科学研究理事会;
关键词
privacy; soft biometrics; disentangled representation learning; variational autoencoder; IDENTIFICATION;
D O I
10.21437/Interspeech.2021-2163
中图分类号
R36 [病理学]; R76 [耳鼻咽喉科学];
学科分类号
100104 ; 100213 ;
摘要
Besides its linguistic content, our speech is rich in biometric information that can be inferred by classifiers. Learning privacy-preserving representations for speech signals enables downstream tasks without sharing unnecessary, private information about an individual. In this paper, we show that protecting gender information in speech is more effective than modelling speaker-identity information only when generating a nonsensitive representation of speech. Our method relies on reconstructing speech by decoding linguistic content along with gender information using a variational autoencoder. Specifically, we exploit disentangled representation learning to encode information about different attributes into separate subspaces that can be factorised independently. We present a novel way to encode gender information and disentangle two sensitive biometric identifiers, namely gender and identity, in a privacyprotecting setting. Experiments on the LibriSpeech dataset show that gender recognition and speaker verification can be reduced to a random guess, protecting against classification-based attacks.
引用
收藏
页码:1699 / 1703
页数:5
相关论文
共 50 条
  • [21] Gender identity and its influence on the speech production
    Ekaterina, Oshchepkova S.
    [J]. FILOLOGICHESKIE NAUKI-NAUCHNYE DOKLADY VYSSHEI SHKOLY-PHILOLOGICAL SCIENCES-SCIENTIFIC ESSAYS OF HIGHER EDUCATION, 2013, (01): : 41 - 53
  • [22] Protecting Identity Violence and Its Representations in France, 1815-1830
    Hage, Ralph
    [J]. CONTAGION-JOURNAL OF VIOLENCE MIMESIS AND CULTURE, 2018, 25 : 49 - 77
  • [23] REPRESENTATIONS ON THE SPEECH GENDER WRITING IN THE PORTUGUESE LANGUAGE CLASSROOM
    Nogueira, Susana dos Santos
    de Lima, Lucielena Mendonca
    [J]. HUMANIDADES & INOVACAO, 2018, 5 (10): : 116 - 131
  • [24] Domain Agnostic Learning with Disentangled Representations
    Peng, Xingchao
    Huang, Zijun
    Sun, Ximeng
    Saenko, Kate
    [J]. INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 97, 2019, 97
  • [25] Learning Disentangled Representations of Negation and Uncertainty
    Vasilakes, Jake
    Zerva, Chrysoula
    Miwa, Makoto
    Ananiadou, Sophia
    [J]. PROCEEDINGS OF THE 60TH ANNUAL MEETING OF THE ASSOCIATION FOR COMPUTATIONAL LINGUISTICS (ACL 2022), VOL 1: (LONG PAPERS), 2022, : 8380 - 8397
  • [26] A Contrastive Objective for Learning Disentangled Representations
    Kahana, Jonathan
    Hoshen, Yedid
    [J]. COMPUTER VISION, ECCV 2022, PT XXVI, 2022, 13686 : 579 - 595
  • [27] Deep Disentangled Representations for Volumetric Reconstruction
    Grant, Edward
    Kohli, Pushmeet
    van Gerven, Marcel
    [J]. COMPUTER VISION - ECCV 2016 WORKSHOPS, PT III, 2016, 9915 : 266 - 279
  • [28] An Identifiable Double VAE For Disentangled Representations
    Mita, Graziano
    Filippone, Maurizio
    Michiardi, Pietro
    [J]. INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 139, 2021, 139
  • [29] Adversarial Robustness through Disentangled Representations
    Yang, Shuo
    Guo, Tianyu
    Wang, Yunhe
    Xu, Chang
    [J]. THIRTY-FIFTH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE, THIRTY-THIRD CONFERENCE ON INNOVATIVE APPLICATIONS OF ARTIFICIAL INTELLIGENCE AND THE ELEVENTH SYMPOSIUM ON EDUCATIONAL ADVANCES IN ARTIFICIAL INTELLIGENCE, 2021, 35 : 3145 - 3153
  • [30] Learning disentangled representations in the imaging domain
    Liu, Xiao
    Sanchez, Pedro
    Thermos, Spyridon
    O'Neil, Alison Q.
    Tsaftaris, Sotirios A.
    [J]. MEDICAL IMAGE ANALYSIS, 2022, 80