AASIST: Audio Anti-Spoofing using Integrated Spectro-Temporal Graph Attention Networks
This addresses the need for efficient and broad-spectrum audio anti-spoofing systems, which is incremental as it builds on existing graph attention methods but introduces novel components for improved performance.
The paper tackled the problem of detecting spoofed audio utterances by developing an efficient single system that avoids computationally demanding ensembles, achieving a 20% relative improvement over the state-of-the-art and with a lightweight variant outperforming all competitors using only 85K parameters.
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.