What You Train Is What You Get: Gender Bias, Training Composition, and Post-Hoc Mitigation in Audio Deepfake Detection
Authors: Aishwarya R. Fursule, Vamshi Nallaguntla, Shruti Kshirsagar, Anderson R. Avila
Abstract
Audio deepfake detection models determine whether speech is genuine or artificially generated, but high overall accuracy can mask substantial performance disparities across demographic groups. In this work, we investigate gender bias in audio deepfake detection using the ASVspoof5 dataset. We use ASVspoof5 under a controlled custom split designed to isolate gender-composition effects. We train attack-specific models on nine training sets with different gender compositions, ranging from female-only to male-only. We use a ResNet18 classifier with LogSpectrogram and WavLM-Base+ features, and we evaluated six post-hoc threshold calibration methods. Experimental results show that training data composition strongly predicts bias direction, with the underrepresented gender performing worse at test time. WavLM-Base+ features are shown to produce gender performance gaps 3.0 to 4.3 times larger than LogSpectrogram under identical training conditions, and balanced training is found to reduce LogSpectrogram bias but leave WavLM bias largely intact. Moreover, all six calibration strategies, including Oracle calibration with full test-set label access, leave the Equal Error Rate gap unchanged at 1.317 pp, confirming that threshold adjustment cannot correct underlying score distribution disparities. Overall, these findings suggest that gender fairness in audio deepfake detection must be addressed at training time, as post-hoc methods can only partially mitigate the resulting disparities
Audio deepfake detection systems are increasingly deployed in high-stakes security applications, yet their fairness across demographic groups remains critically underexamined. Prior work measures gender disparity but does not investigate where it comes from or how to fix it systematically. We present the first diagnosis-first framework that identifies bias source before applying targeted mitigation, evaluated on two models, AASIST and Wav2Vec2+ResNet18, on ASVSpoof5. Our diagnosis shows that bias does not stem from imbalanced training data but from acoustic representation differences, gender leakage in learned features, and structural evaluation asymmetry. We test mitigation strategies across in-processing, post-processing and combined families, including novel methods introduced in this work. Adjusting the decision threshold separately per gender reduces unfairness by 54% to 75% at no cost to detection accuracy, and our new epoch-level fairness regularisation method outperforms existing per-batch approaches. Adversarial debiasing succeeds only when gender leakage is localised, and fails when it is diffuse, an outcome correctly predicted by our diagnosis before training. No single method fully closes the fairness gap, confirming that bias sources must be identified before fixes are applied and that fairer benchmark design is equally important
Aishwarya Fursule, Shruti Kshirsagar, Anderson R. Avila
Audio deepfake detectors often fail to generalize across speakers, as they learn speaker-identity features rather than synthesis artifacts, known as implicit identity leakage. Existing methods address this but incur architectural complexity or training instability. This paper proposes a dual-granularity orthogonal disentanglement framework enforcing feature independence at two levels: sample-level cosine orthogonality captures directional decorrelation, while batch-level cross-covariance regularization eliminates linear correlations across embedding dimensions. A curriculum disentanglement schedule progressively strengthens the orthogonality constraint without auxiliary networks or adversarial dynamics. Experiments on ASVspoof 2019 LA, ASVspoof 2021 DF, and In-the-Wild datasets demonstrate that the proposed method achieves 1.35%, 7.88%, and 21.58% equal error rates (EER), respectively, surpassing gradient reversal disentanglement by 2.60% absolute on cross-dataset transfer.
Claims about the robustness and fairness of deepfake speech detectors are only as credible as the datasets used to train and evaluate those systems. We present a dataset-level audit of the deepfake speech landscape. We compile and analyze 39 deepfake speech datasets, examining key attributes including accessibility, documentation, demographic and language coverage, dataset scale, and the underlying bona fide speech sources. Our audit reveals two important takeaways. Firstly, fairness assessment is largely infeasible because most datasets lack demographic metadata, and only a few contain gender or language labels. This prevents any meaningful subgroup analysis and leaves other demographic attributes unaddressed. Secondly, we identify substantial overlap in underlying bona fide source corpora across datasets, which can undermine cross-dataset evaluation and lead to overstated generalization claims.