Meta-Learning Approaches for Improving Detection of Unseen Speech Deepfakes

TL;DR

This paper explores meta-learning techniques to enhance the detection of unseen speech deepfakes, achieving significant improvements in generalization with minimal samples, addressing a key challenge in real-world applications.

Contribution

The study introduces a meta-learning framework for attack-invariant feature learning, enabling effective few-shot adaptation to unseen speech deepfake attacks.

Findings

EER reduced from 21.67% to 10.42% on InTheWild dataset

Significant improvement with only 96 samples from unseen attacks

Continuous adaptation maintains detection performance over time

Abstract

Current speech deepfake detection approaches perform satisfactorily against known adversaries; however, generalization to unseen attacks remains an open challenge. The proliferation of speech deepfakes on social media underscores the need for systems that can generalize to unseen attacks not observed during training. We address this problem from the perspective of meta-learning, aiming to learn attack-invariant features to adapt to unseen attacks with very few samples available. This approach is promising since generating of a high-scale training dataset is often expensive or infeasible. Our experiments demonstrated an improvement in the Equal Error Rate (EER) from 21.67% to 10.42% on the InTheWild dataset, using just 96 samples from the unseen dataset. Continuous few-shot adaptation ensures that the system remains up-to-date.