# A Probe Shift Is Not a Fairness Fix: The Limits of Representation Steering in Speech Models

arXiv: https://arxiv.org/abs/2609.18533 · HTML: https://mlsanity.com/paper/2609.18533/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in SAE features and their applications, one of 13 directions in Mechanistic interpretability.
- Builds on 19 of its 26 references that are in the map.
- First posted on arXiv 16 Sep 2026.

## Placed in

- Mechanistic interpretability / [SAE features and their applications](https://mlsanity.com/mechinterp/sae-features-and-their-applications/)

## Builds on (most cited first, 19 of 19)

- [Robust Speech Recognition via Large-Scale Weak Supervision](https://mlsanity.com/paper/2212.04356/) (2022)
- [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://mlsanity.com/paper/2006.11477/) (2020)
- [HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units](https://mlsanity.com/paper/2106.07447/) (2021)
- [Common Voice: A Massively-Multilingual Speech Corpus](https://mlsanity.com/paper/1912.06670/) (2019)
- [Understanding intermediate layers using linear classifier probes](https://mlsanity.com/paper/1610.01644/) (2016)
- [Steering Llama 2 via Contrastive Activation Addition](https://mlsanity.com/paper/2312.06681/) (2023)
- [Probing Classifiers: Promises, Shortcomings, and Advances](https://mlsanity.com/paper/2102.12452/) (2021)
- [The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasets](https://mlsanity.com/paper/2310.06824/) (2023)
- [Layer-wise Analysis of a Self-supervised Speech Representation Model](https://mlsanity.com/paper/2107.04734/) (2021)
- [Quantifying Bias in Automatic Speech Recognition](https://mlsanity.com/paper/2103.15122/) (2021)
- [Probing Classifiers are Unreliable for Concept Removal and Detection](https://mlsanity.com/paper/2207.04153/) (2022)
- [Analyzing Hidden Representations in End-to-End Automatic Speech Recognition Systems](https://mlsanity.com/paper/1709.04482/) (2017)
- [To Steer or Not to Steer? Mechanistic Error Reduction with Abstention for Language Models](https://mlsanity.com/paper/2510.13290/) (2025)
- [Twists, Humps, and Pebbles: Multilingual Speech Recognition Models Exhibit Gender Performance Gaps](https://mlsanity.com/paper/2402.17954/) (2024)
- [On the social bias of speech self-supervised models](https://mlsanity.com/paper/2406.04997/) (2024)
- [Adversarial Disentanglement of Speaker Representation for Attribute-Driven Privacy Preservation](https://mlsanity.com/paper/2012.04454/) (2020)
- [Unsupervised Speaker Adaptation using Attention-based Speaker Memory for End-to-End ASR](https://mlsanity.com/paper/2002.06165/) (2020)
- [Identifying and typifying demographic unfairness in phoneme-level embeddings of self-supervised speech recognition models](https://mlsanity.com/paper/2604.22631/) (2026)
- [Speaker Group Encoding in Self-supervised Speech Recognition Models](https://mlsanity.com/paper/2606.10654/) (2026)

## Abstract (arXiv)

Automatic speech recognition (ASR) systems exhibit unequal error rates across speaker groups, motivating interventions on their internal representations. We ask whether speaker-linked attributes that are linearly readable from pretrained ASR encoders yield useful directions for reducing group word-error-rate (WER) gaps. Across Whisper-medium, HuBERT-large, and Wav2Vec2-large on Common Voice and the Speech Accent Archive, we probe every encoder layer for metadata-derived sex/gender, age, and native/accent labels; construct centroid and probe-derived directions; inject them at selected layers; and compare downstream probe trajectories with matched WER changes. Sex labels are highly decodable (best macro-F1 0.924--0.941), native/accent labels are also above chance (0.544--0.696), and age is weaker (0.354--0.397). Of 22 post-selected reruns, nine have 95% paired-bootstrap intervals entirely below zero, yet every absolute source-group WER reduction is below 0.7 percentage points. Conversely, a local target-class probe rate can rise from 8.09% to 99.87% while WER worsens. Linear readability is therefore neither evidence of causal use nor a reliable mitigation method. Our results motivate evaluating speech-bias interventions jointly at representation, propagation, and task levels.