SpAArSIST: Sparsified AASIST for Efficient and Reliable Anti-Spoofing

2026-06-10

AI Summary

This work addresses the computational redundancy, large model size, and poor out-of-domain generalization inherent in existing self-supervised anti-spoofing systems by proposing the first deployment-oriented sparsification scheme for AASIST. By replacing learnable graph pooling and stacked node attention mechanisms with an explicitly lightweight design—featuring decoupled training/inference graph pooling ratios, magnitude-based node scoring, and mean aggregation—the proposed approach significantly reduces computational overhead by 20.7% and model parameters by 4.1%, while simultaneously enhancing out-of-domain robustness. The method achieves an EER of 2.82% and a minDCF of 0.078 on the In-the-Wild dataset, and maintains strong performance on ASVspoof5, demonstrating its effectiveness and practicality for real-world deployment.

Abstract

We present SpAArSIST, a deployment-oriented refinement of the widely used AASIST graph pooling backend for self-supervised learning (SSL) based anti-spoofing. Motivated by redundant operations in public implementations, we replace learned pooling and stack-node attention with explicit, lightweight choices: separate train and inference graph pooling ratios $(k_{\mathrm{tr}},k_{\mathrm{inf}})$, magnitude-based node scoring, and mean aggregation of graph nodes. The best overall configuration (rank 1) cuts backend compute by 20.7% (195.045M $\rightarrow$ 154.706M MACs) and model size by 4.1% (611.8k $\rightarrow$ 586.4k params), while improving out-of-domain robustness on In-the-Wild to 2.82% EER and 0.078 minDCF (from 4.64% and 0.133) and remaining competitive on ASVspoof5. We further provide a composite selection score that summarizes accuracy, calibration, and compute to support balanced deployment-oriented model choice.

Problem

• anti-spoofing
• self-supervised learning
• graph pooling
• model efficiency
• out-of-domain robustness

Innovation

• Sparsified Graph Pooling
• Anti-Spoofing
• Self-Supervised Learning
• Efficient Inference
• Out-of-Domain Robustness