One-Class Learning with Adaptive Centroid Shift for Audio Deepfake Detection

TL;DR

This paper introduces an adaptive centroid shift method for one-class learning that improves deepfake audio detection by creating robust, well-separated embeddings of bonafide speech, outperforming existing systems.

Contribution

The paper proposes a novel adaptive centroid shift technique that enhances one-class learning for audio deepfake detection, focusing solely on bonafide samples for improved robustness.

Findings

Achieves 2.19% EER on ASVspoof 2021 dataset

Effectively maps bonafide embeddings into a single cluster

Successfully disentangles bonafide and spoof classes

Abstract

As speech synthesis systems continue to make remarkable advances in recent years, the importance of robust deepfake detection systems that perform well in unseen systems has grown. In this paper, we propose a novel adaptive centroid shift (ACS) method that updates the centroid representation by continually shifting as the weighted average of bonafide representations. Our approach uses only bonafide samples to define their centroid, which can yield a specialized centroid for one-class learning. Integrating our ACS with one-class learning gathers bonafide representations into a single cluster, forming well-separated embeddings robust to unseen spoofing attacks. Our proposed method achieves an equal error rate (EER) of 2.19% on the ASVspoof 2021 deepfake dataset, outperforming all existing systems. Furthermore, the t-SNE visualization illustrates that our method effectively maps the bonafide embeddings into a single cluster and successfully disentangles the bonafide and spoof classes.