From Dispersion to Attraction: Spectral Dynamics of Hallucination Across Whisper Model Scales

View PDF HTML (experimental) Abstract:Hallucinations in large ASR models present a critical safety risk. In this work, we propose the \textit{Spectral Sensitivity Theorem}, which predicts a phase transition in deep networks from a dispersive regime (signal decay) to an attractor regime (rank-1 collapse) governed by layer-wise gain and alignment. We validate this theory by analyzing the eigenspectra of activation graphs in Whisper models (Tiny to Large-v3-Turbo) under adversarial stress. Our results confirm the theoretical prediction: intermediate models exhibit \textit{Structural Disintegration} (Regime I), characterized by a $13.4\%$ collapse in Cross-Attention rank. Conversely, large models enter a \textit{Compression-Seeking Attractor} state (Regime II), where Self-Attention actively compresses rank ($-2.34\%$) and hardens the spectral slope, decoupling the model from acoustic evidence. Comments: This paper has been submitted to Interspeech 2026 for review Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI) Cite as: arXiv:2604.08591 [cs.LG] (or arXiv:2604.08591v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2604.08591 arXiv-issued DOI via DataCite Submission history From: Kirill Borodin [view email] [v1] Tue, 31 Mar 2026 22:17:30 UTC (902 KB)