Detection of voice conversion spoofing attacks using voiced speech
Loading...
Files
Date
2022-11-11
Authors
Sankar, M. S. Arun
De Leon, Phillip L.
Roedig, Utz
Journal Title
Journal ISSN
Volume Title
Publisher
Springer
Published Version
Abstract
Speech consists of voiced and unvoiced segments that differ in their production process and exhibit different characteristics. In this paper, we investigate the spectral differences between bonafide and spoofed speech for voiced and unvoiced speech segments. We observe that the largest spectral differences lie in the 0–4 kHz band of voiced speech. Based on this observation, we propose a low-complexity, pre-processing stage which subsamples voiced frames prior to spoofing detection. The proposed pre-processing stage is applied to two systems, LFCC+GMM and IA/IF+KNN that differ entirely on the features and classifier used for spoofing detection. Our results show improvement with both systems in detection of the ASVspoof 2019 A17 voice conversion attack, which is recognized to have one of the highest spoofing capabilities. We also show improvements in the A18 and A19 voice conversion attacks for the IA/IF+KNN system. The resulting A17 EERs are lower than all reported systems where the A17 spoofing attack is the worst attack except the Capsule Network. Finally, we note that the proposed pre-processing stage reduces the speech date by more than 4× due to subsampling and using only voiced frames but at the same time maintaining similar pooled EER as that for the baseline systems, which may be advantageous for resource constrained spoofing detectors.
Description
Keywords
Spoofing detection , Speech processing , Computer security , Voice bio-metric
Citation
Sankar, Arun M. S.., L. De Leon, P. and Roedig, U. (2022) ‘Detection of voice conversion spoofing attacks using voiced speech’, NordSec 2022, 27th Nordic Conference on Secure IT Systems, Reykjavik, Iceland, 30 Nov-02 Dec, in H.P. Reiser and M. Kyas (eds) Secure IT Systems, Lecture Notes in Computer Science, vol 13700, Cham: Springer International Publishing, pp. 159-175. https://doi.org/10.1007/978-3-031-22295-5_9
Link to publisher’s version
Collections
Copyright
© 2022 The Author(s). This is a post-peer-review, pre-copyedit version of an article published in Lecture Notes in Computer Science. The final authenticated version is available online at: https://doi.org/10.1007/978-3-031-22295-5_9