Development of generalizable countermeasures for replay spoofing attacks on Automatic Speaker Verification (ASV) systems is still an open problem. Many countermeasures to date utilize bandpass filters to extract a variety of frequency band-based features. This paper proposes the use of adaptive bandpass filters, a concept adopted from human cochlear modelling to improve detection performance. Gains of filters used for subband based feature extraction are adaptively adjusted by varying their Q factors (Quality factor) as a function of input signal level to boost low amplitude signal components and improve the front-end’s sensitivity to them. This method is used to enhance information embedded in speech signals such as device channel effects which could be instrumental in distinguishing genuine speech signals from replayed ones. Three features extracted using the adaptive filter process yielded performance improvements over other auditory concepts-based baselines, showing the potential of using an adaptive filter mechanism for replay spoofing attack detection.
Cite as: Wickramasinghe, B., Ambikairajah, E., Epps, J. (2019) Biologically Inspired Adaptive-Q Filterbanks for Replay Spoofing Attack Detection. Proc. Interspeech 2019, 2953-2957, doi: 10.21437/Interspeech.2019-1535
@inproceedings{wickramasinghe19_interspeech,
author={Buddhi Wickramasinghe and Eliathamby Ambikairajah and Julien Epps},
title={{Biologically Inspired Adaptive-Q Filterbanks for Replay Spoofing Attack Detection}},
year=2019,
booktitle={Proc. Interspeech 2019},
pages={2953--2957},
doi={10.21437/Interspeech.2019-1535}
}