In this work, we address the challenge of generalizable audio deepfake detection (ADD) across diverse speech synthesis paradigms-including conventional text-to-speech (TTS) systems and modern diffusion or flow-matching (FM) based generators. Prior work has mostly targeted individual synthesis families and often fails to generalize across paradigms due to overfitting to generation-specific artifacts. We hypothesize that synthetic speech, irrespective of its generative origin, leaves behind shared structural distortions in the embedding space that can be aligned through geometry-aware modeling. To this end, we propose RHYME, a unified detection framework that fuses utterance-level embeddings from diverse pretrained speech encoders using non-Euclidean projections. RHYME maps representations into hyperbolic and spherical manifolds-where hyperbolic geometry excels at modeling hierarchical generator families, and spherical projections capture angular, energy-invariant cues such as periodic vocoder artifacts. The fused representation is obtained via Riemannian barycentric averaging, enabling synthesis-invariant alignment. RHYME outperforms individual PTMs and homogeneous fusion baselines, achieving top performance and setting new state-of-the-art in cross-paradigm ADD.
19 June 2026 in ACM Transactions on Computing for Healthcare
In this study, we introduce Uc-PrUn , a principled framework designed to improve the reliability of Vision–Language Models (VLMs) in clinical decision-support systems. Recognizing the critical importance of uncertainty estimation in high-stakes domains, we introduce a two-stage methodology; the first stage focuses on Bayesian-inspired zero-shot uncertainty quantification using Monte Carlo dropout, while the second stage introduces a novel uncertainty-aware machine unlearning strategy, termed Uc-PrUn . Leveraging the Harvard-FairVLMed dataset, which comprises paired SLO fundus images and clinical notes for glaucoma detection, we systematically evaluate various VLMs to quantify epistemic uncertainty and identify high-variance training samples. Our pruning and unlearning mechanism selectively removes uncertain samples to enhance model calibration and improve downstream performance. Experimental results demonstrate that the Uc-PrUn approach not only reduces predictive uncertainty but also yields consistent gains in accuracy and F 1 -scores across multiple Vision–Language Models. These findings support incorporating uncertainty-aware pruning mechanisms in medical AI pipelines where model reliability is essential.