AASIST: Audio anti-spoofing using integrated spectro-temporal graph attention networks

Jung, Jee-weon; Heo, Hee-Soo; Tak, Hemlata; Shim, Hye-jin; Chung, Joon Son; Lee, Bong-Jin; Yu, Ha-Jin; Evans, Nicholas
Submitted to ArXiV, 4 October 2021

Artefacts that differentiate spoofed from bona-fide utterances can reside in spectral or temporal domains. Their reliable detection usually depends upon computationally demanding ensemble systems where each subsystem is tuned to some specific artefacts. We seek to develop an efficient, single system that can detect a broad range of different spoofing attacks without score-level ensembles. We propose a novel heterogeneous stacking graph attention layer which models artefacts spanning heterogeneous temporal and spectral domains with a heterogeneous attention mechanism and a stack node. With a new max graph operation that involves a competitive mechanism and an extended readout scheme, our approach, named AASIST, outperforms the current state-of-the-art by 20% relative. Even a lightweight variant, AASIST-L, with only 85K parameters, outperforms all competing systems.


Type:
Conference
Date:
2021-10-04
Department:
Digital Security
Eurecom Ref:
6696
Copyright:
© EURECOM. Personal use of this material is permitted. The definitive version of this paper was published in Submitted to ArXiV, 4 October 2021 and is available at :

PERMALINK : https://www.eurecom.fr/publication/6696