Replay attacks pose the most severe threat to automatic speaker verification systems amongvarious spoofing attacks. In this paper, we propose a novel feature extraction method thatleverages both the phase-based and magnitude-based features. The proposed method fullyutilizes the subband information and the complementary information from the phase andmagnitude spectra. First, we conduct a discriminative performance analysis on full frequencybands via the F-ratio method. Then, variable-frequency resolution features are extracted viaseveral techniques to capture highly discriminative information on frequency bands. Finally,complementary information from the phase and magnitude domains are fused to achievehigher performance. The results on the ASVspoof 2017 database demonstrate that our pro-posed frequency adaptive features attain relative error reduction rates of 83.4% and 62.3% onthe development and evaluation datasets, respectively, compared to the baseline method,
License type:
http://creativecommons.org/licenses/by-nc-nd/4.0/
Funding Info:
The research was supported partially by the National Key R&D Program of China (2018YFB1305200), National Natural Science Foundation of China (61771333), TianjinMunicipal Science and Technology Project (18ZXZNGX00330) and JSPS KAKENHI Grant (No.16K12461). There is no ASTAR and local funding involved in this paper.
Description:
Please find the link to the article at the publisher's URL: https://doi.org/10.1016/j.csl.2020.101161