# Speech is Silver, Silence is Golden: What do ASVspoof-trained Models Really Learn?

arXiv: https://arxiv.org/abs/2106.12914 · HTML: https://mlsanity.com/paper/2106.12914/

## Key facts

- Cited by 37 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 23 Jun 2021.

## Built on by (most cited first, 30 of 37)

- [Does Audio Deepfake Detection Generalize?](https://mlsanity.com/paper/2203.16263/) (2022)
- [Circumventing shortcuts in audio-visual deepfake detection datasets with unsupervised learning](https://mlsanity.com/paper/2412.00175/) (2024)
- [Human Perception of Audio Deepfakes](https://mlsanity.com/paper/2107.09667/) (2021)
- [How Do Neural Spoofing Countermeasures Detect Partially Spoofed Audio?](https://mlsanity.com/paper/2406.02483/) (2024)
- [Harder or Different? Understanding Generalization of Audio Deepfake Detection](https://mlsanity.com/paper/2406.03512/) (2024)
- [SLIM: Style-Linguistics Mismatch Model for Generalized Audio Deepfake Detection](https://mlsanity.com/paper/2407.18517/) (2024)
- [Complex-valued neural networks for voice anti-spoofing](https://mlsanity.com/paper/2308.11800/) (2023)
- [Collaborative Watermarking for Adversarial Speech Synthesis](https://mlsanity.com/paper/2309.15224/) (2023)
- [Attacker Attribution of Audio Deepfakes](https://mlsanity.com/paper/2203.15563/) (2022)
- [DSVAE: Interpretable Disentangled Representation for Synthetic Speech Detection](https://mlsanity.com/paper/2304.03323/) (2023)
- [How to Construct Perfect and Worse-than-Coin-Flip Spoofing Countermeasures: A Word of Warning on Shortcut Learning](https://mlsanity.com/paper/2306.00044/) (2023)
- [Replay Attacks Against Audio Deepfake Detection](https://mlsanity.com/paper/2505.14862/) (2025)
- [Shortcut Learning in Binary Classifier Black Boxes: Applications to Voice Anti-Spoofing and Biometrics](https://mlsanity.com/paper/2601.17782/) (2026)
- [The Sound of Silence: Efficiency of First Digit Features in Synthetic Audio Detection](https://mlsanity.com/paper/2210.02746/) (2022)
- [Shortcut Detection with Variational Autoencoders](https://mlsanity.com/paper/2302.04246/) (2023)
- [Dual-Branch Knowledge Distillation for Noise-Robust Synthetic Speech Detection](https://mlsanity.com/paper/2310.08869/) (2023)
- [Easy, Interpretable, Effective: openSMILE for voice deepfake detection](https://mlsanity.com/paper/2408.15775/) (2024)
- [Audio Codec Augmentation for Robust Collaborative Watermarking of Speech Synthesis](https://mlsanity.com/paper/2409.13382/) (2024)
- [What Does an Audio Deepfake Detector Focus on? A Study in the Time Domain](https://mlsanity.com/paper/2501.13887/) (2025)
- [Beyond Silence: Bias Analysis through Loss and Asymmetric Approach in Audio Anti-Spoofing](https://mlsanity.com/paper/2406.17246/) (2024)
- [Optimizing Tandem Speaker Verification and Anti-Spoofing Systems](https://mlsanity.com/paper/2201.09709/) (2022)
- [Localized Shortcut Removal](https://mlsanity.com/paper/2211.15510/) (2022)
- [Compression Robust Synthetic Speech Detection Using Patched Spectrogram Transformer](https://mlsanity.com/paper/2402.14205/) (2024)
- [DeePen: Penetration Testing for Audio Deepfake Detection](https://mlsanity.com/paper/2502.20427/) (2025)
- [On Deepfake Voice Detection -- It's All in the Presentation](https://mlsanity.com/paper/2509.26471/) (2025)
- [Enhancing Multi-Corpus Training in SSL-Based Anti-Spoofing Models: Domain-Invariant Feature Extraction](https://mlsanity.com/paper/2603.18657/) (2026)
- [Alethia: A Foundational Encoder for Voice Deepfakes](https://mlsanity.com/paper/2605.00251/) (2026)
- [The Watermark Shortcut: How Provenance Marking Sabotages Audio Deepfake Detection](https://mlsanity.com/paper/2606.23335/) (2026)
- [An Intervention-Based Framework for Shortcut Diagnosis in Spoofing Countermeasures](https://mlsanity.com/paper/2607.03150/) (2026)
- [Protecting Publicly Available Data With Machine Learning Shortcuts](https://mlsanity.com/paper/2310.19381/) (2023)

## Abstract (arXiv)

We present our analysis of a significant data artifact in the official 2019/2021 ASVspoof Challenge Dataset. We identify an uneven distribution of silence duration in the training and test splits, which tends to correlate with the target prediction label. Bonafide instances tend to have significantly longer leading and trailing silences than spoofed instances. In this paper, we explore this phenomenon and its impact in depth. We compare several types of models trained on a) only the duration of the leading silence and b) only on the duration of leading and trailing silence. Results show that models trained on only the duration of the leading silence perform particularly well, and achieve up to 85% percent accuracy and an equal error rate (EER) of 15.1%. At the same time, we observe that trimming silence during pre-processing and then training established antispoofing models using signal-based features leads to comparatively worse performance. In that case, EER increases from 3.6% (with silence) to 15.5% (trimmed silence). Our findings suggest that previous work may, in part, have inadvertently learned thespoof/bonafide distinction by relying on the duration of silence as it appears in the official challenge dataset. We discuss the potential consequences that this has for interpreting system scores in the challenge and discuss how the ASV community may further consider this issue.