Frame-level Temporal Difference Learning for Partial Deepfake Speech Detection

TL;DR

This paper introduces a novel deepfake speech detection method that analyzes frame-level temporal differences to identify unnatural variations, achieving state-of-the-art results without needing costly frame-level annotations.

Contribution

It proposes a Temporal Difference Attention Module (TDAM) that detects partial deepfakes by modeling temporal irregularities at multiple scales without explicit boundary labels.

Findings

Achieves an EER of 0.59% on PartialSpoof dataset

Achieves an EER of 0.03% on HAD dataset

Outperforms existing methods significantly

Abstract

Detecting partial deepfake speech is essential due to its potential for subtle misinformation. However, existing methods depend on costly frame-level annotations during training, limiting real-world scalability. Also, they focus on detecting transition artifacts between bonafide and deepfake segments. As deepfake generation techniques increasingly smooth these transitions, detection has become more challenging. To address this, our work introduces a new perspective by analyzing frame-level temporal differences and reveals that deepfake speech exhibits erratic directional changes and unnatural local transitions compared to bonafide speech. Based on this finding, we propose a Temporal Difference Attention Module (TDAM) that redefines partial deepfake detection as identifying unnatural temporal variations, without relying on explicit boundary annotations. A dual-level hierarchical difference representation captures temporal irregularities at both fine and coarse scales, while adaptive average pooling preserves essential patterns across variable-length inputs to minimize information loss. Our TDAM-AvgPool model achieves state-of-the-art performance, with an EER of 0.59% on the PartialSpoof dataset and 0.03% on the HAD dataset, which significantly outperforms the existing methods without requiring frame-level supervision.