SOURCE-LINKED INTELLIGENCE

AURA: Uncertainty-Routed Activation Editing for Acoustic Grounding in Speech Foundation Models

arXiv · AI, language, vision and robotics · article · Sep 21, 2026 · UTC

Attention encoder-decoder (AED) Speech Foundation Models achieve strong ASR performance but can generate acoustically unsupported text when inputs contain no speech, weak acoustic evidence, or unreliable transcription. We propose AURA: Activation-editing with Uncertainty-Routed Adaptation, an ultra-efficient representation-editing method that freezes the pretrained model and applies sparse scale-and-shift edits to decoder cross-attention heads. AURA dynamically routes edits using cross-attention uncertainty features that capture over-concentration, diffuse attention, and abrupt frame shifts. W

Read original source ↗ Open in workspace

recordType

paper

region

Global