Searcharxiv⌕ Search
Home/Archive/arXiv

arXiv · 2507.15101

Frame-level Temporal Difference Learning for Partial Deepfake Speech Detection

Menglu Li·Xiao-Ping Zhang·Lian Zhao
View original source ↗Read source document ↗Export metadata ↓

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.

Explore related subjects

cs.SDcs.CReess.AS

Keep this discovery

BibTeXRIS

Menglu Li, Xiao-Ping Zhang, Lian Zhao. 2025-07-20. Frame-level Temporal Difference Learning for Partial Deepfake Speech Detection. https://doi.org/10.1109/lsp.2025.3592111

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Xiaomi-CocktailASR-1 Technical Report

Recently, large language model (LLM) based ASR models have achieved significant progress, yet they generally lack support for multi-speaker scenarios, where the cocktail party problem remains a critical bottleneck for further advancing ASR. Existing TS-ASR methods, including end-to-end architectures with speaker embeddings and latest LLM-based explorations suffer from degraded single-speaker performance and the inability to reject when the target speaker is absent. In this paper, we propose Xiaomi-CocktailASR-1, an LLM-based end-to-end TS-ASR architecture. By utilizing reference speech as voiceprint prompts, it directly transcribes the target speaker's speech without requiring speech separation. Xiaomi-CocktailASR-1 maintains competitive performance in single-speaker scenarios, comparable to mainstream ASR models. It also features a negative sample rejection capability, outputting empty text when the target speaker is absent from the mixed speech. Additionally, Xiaomi-CocktailASR-1 supports a Chain-of-Thought (CoT) reasoning mode to provide explicit reasoning steps. Extensive experiments on various synthetic and real-world multispeaker benchmarks demonstrate that Xiaomi-CocktailASR-1 achieves state-of-the-art performance, effectively addressing the cocktail party problem through a unified architecture that balances multispeaker and single-speaker recognition accuracy, along with rejection capability.

cs.SD↗
arXiv2026

EConv-TasNet: Efficient Conv-TasNet for Effective Speech Separation

Conv-TasNet has served as a strong baseline for time-domain speech separation, and many studies have extended it with advanced architectures such as dual-path networks, U-Nets, and attention mechanisms. However, these methods often introduce high computational cost and complexity, limiting their deployment in resource-constrained scenarios. To address this issue, we propose eConv-TasNet, an efficient variant of Conv-TasNet that improves both effectiveness and efficiency without relying on resource-intensive modules. The proposed model consists of a group-wise early-splitting (GES) module and a multi-group feature aggregation (MGFA) module. GES generates discriminative speaker embeddings at intermediate stages, while MGFA progressively aggregates these group-level representations for refined mask estimation. Experimental results show that eConv-TasNet reduces model size by 22.4%, accelerates inference by 18.9%, and improves SI-SNRi by 14.0%-28.0% across three public benchmarks. Moreover, it achieves competitive performance compared with state-of-the-art methods while requiring significantly fewer parameters and lower inference cost. These results demonstrate a favorable efficiency-effectiveness trade-off for edge deployment.

cs.SD↗