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

arXiv: https://arxiv.org/abs/2609.23979 · HTML: https://mlsanity.com/paper/2609.23979/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 36 of its 48 references that are in the map.
- First posted on arXiv 21 Sep 2026.

## Builds on (most cited first, 30 of 36)

- [LoRA: Low-Rank Adaptation of Large Language Models](https://mlsanity.com/paper/2106.09685/) (2021)
- [Parameter-Efficient Transfer Learning for NLP](https://mlsanity.com/paper/1902.00751/) (2019)
- [Robust Speech Recognition via Large-Scale Weak Supervision](https://mlsanity.com/paper/2212.04356/) (2022)
- [BitFit: Simple Parameter-efficient Fine-tuning for Transformer-based Masked Language-models](https://mlsanity.com/paper/2106.10199/) (2021)
- [Analyzing Multi-Head Self-Attention: Specialized Heads Do the Heavy Lifting, the Rest Can Be Pruned](https://mlsanity.com/paper/1905.09418/) (2019)
- [Are Sixteen Heads Really Better than One?](https://mlsanity.com/paper/1905.10650/) (2019)
- [Learning Sparse Neural Networks through L0 Regularization](https://mlsanity.com/paper/1712.01312/) (2017)
- [MUSAN: A Music, Speech, and Noise Corpus](https://mlsanity.com/paper/1510.08484/) (2015)
- [Joint CTC-Attention based End-to-End Speech Recognition using Multi-task Learning](https://mlsanity.com/paper/1609.06773/) (2016)
- [TED-LIUM 3: twice as much data and corpus repartition for experiments on speaker adaptation](https://mlsanity.com/paper/1805.04699/) (2018)
- [ReFT: Representation Finetuning for Language Models](https://mlsanity.com/paper/2404.03592/) (2024)
- [OWSM v3.1: Better and Faster Open Whisper-Style Speech Models based on E-Branchformer](https://mlsanity.com/paper/2401.16658/) (2024)
- [LoFiT: Localized Fine-tuning on LLM Representations](https://mlsanity.com/paper/2406.01563/) (2024)
- [Careless Whisper: Speech-to-Text Hallucination Harms](https://mlsanity.com/paper/2402.08021/) (2024)
- [SEP-28k: A Dataset for Stuttering Event Detection From Podcasts With People Who Stutter](https://mlsanity.com/paper/2102.12394/) (2021)
- [Advancing Parameter Efficiency in Fine-tuning via Representation Editing](https://mlsanity.com/paper/2402.15179/) (2024)
- [Investigation of Whisper ASR Hallucinations Induced by Non-Speech Audio](https://mlsanity.com/paper/2501.11378/) (2025)
- [Benchmarking Children's ASR with Supervised and Self-supervised Speech Foundation Models](https://mlsanity.com/paper/2406.10507/) (2024)
- [Detecting Dysfluencies in Stuttering Therapy Using wav2vec 2.0](https://mlsanity.com/paper/2204.03417/) (2022)
- [Kid-Whisper: Towards Bridging the Performance Gap in Automatic Speech Recognition for Children VS. Adults](https://mlsanity.com/paper/2309.07927/) (2023)
- [Lost in Transcription, Found in Distribution Shift: Demystifying Hallucination in Speech Foundation Models](https://mlsanity.com/paper/2502.12414/) (2025)
- [Calm-Whisper: Reduce Whisper Hallucination On Non-Speech By Calming Crazy Heads Down](https://mlsanity.com/paper/2505.12969/) (2025)
- [Joint Localization and Activation Editing for Low-Resource Fine-Tuning](https://mlsanity.com/paper/2502.01179/) (2025)
- [Hallucination Benchmark for Speech Foundation Models](https://mlsanity.com/paper/2510.16567/) (2025)
- [HalluAudio: A Comprehensive Benchmark for Hallucination Detection in Large Audio-Language Models](https://mlsanity.com/paper/2604.19300/) (2026)
- [Whisper Has an Internal Word Aligner](https://mlsanity.com/paper/2509.09987/) (2025)
- [Bottleneck Low-rank Transformers for Low-resource Spoken Language Understanding](https://mlsanity.com/paper/2206.14318/) (2022)
- [Benchmarking Training Paradigms, Dataset Composition, and Model Scaling for Child ASR in ESPnet](https://mlsanity.com/paper/2508.16576/) (2025)
- [Fine-Tuning ASR for Stuttered Speech: Personalized vs. Generalized Approaches](https://mlsanity.com/paper/2506.00853/) (2025)
- [Addressing Pitfalls in Auditing Practices of Automatic Speech Recognition Technologies: A Case Study of People with Aphasia](https://mlsanity.com/paper/2506.08846/) (2025)

## Abstract (arXiv)

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. We evaluate AURA on four datasets spanning non-speech hallucination and speech grounding stressors, including imperfect-label child speech, imperfect-label adult speech, and disfluent speech. On non-speech audio, AURA reduces hallucination rate from 89.18% to 1.94% without prior hallucination-head identification. On imperfect-label corpora, AURA approaches LoRA WER while using roughly 500x fewer trainable parameters. Sensitivity analysis and qualitative cross-attention examples are consistent with AURA's uncertainty-routed editing behavior, supporting dynamic activation editing as a practical path for grounding AED speech models.