Evolutionary Multi-Objective Fusion of Deepfake Speech Detectors

2026-04-01

AI Summary

This work addresses the challenge of improving robustness in deepfake speech detection while avoiding model bloat and diminishing returns commonly associated with existing ensemble fusion methods. It introduces, for the first time, a multi-objective evolutionary optimization approach to this task, proposing an NSGA-II-based score fusion framework that jointly optimizes detection error and system complexity. The study investigates both binary selection and real-valued weighting encoding strategies. Evaluated on the ASVspoof 5 benchmark, the proposed method achieves state-of-the-art performance—EER of 2.37% and minDCF of 0.0684—with approximately half the parameter count of conventional approaches. Furthermore, it yields a Pareto front of solutions that consistently outperforms simple averaging and logistic regression baselines, offering a diverse set of trade-offs between accuracy and model complexity.

Abstract

While deepfake speech detectors built on large self-supervised learning (SSL) models achieve high accuracy, employing standard ensemble fusion to further enhance robustness often results in oversized systems with diminishing returns. To address this, we propose an evolutionary multi-objective score fusion framework that jointly minimizes detection error and system complexity. We explore two encodings optimized by NSGA-II: binary-coded detector selection for score averaging and a real-valued scheme that optimizes detector weights for a weighted sum. Experiments on the ASVspoof 5 dataset with 36 SSL-based detectors show that the obtained Pareto fronts outperform simple averaging and logistic regression baselines. The real-valued variant achieves 2.37% EER (0.0684 minDCF) and identifies configurations that match state-of-the-art performance while significantly reducing system complexity, requiring only half the parameters. Our method also provides a diverse set of trade-off solutions, enabling deployment choices that balance accuracy and computational cost.

Problem

• deepfake speech detection
• ensemble fusion
• system complexity
• multi-objective optimization
• self-supervised learning

Innovation

• evolutionary multi-objective optimization
• deepfake speech detection
• score fusion
• NSGA-II
• self-supervised learning