Paper Overview
Field: NLP Authors: Carolin Holtermann, Minh Duc Bui, Kaitlyn Zhou, Valentin Hofmann, Katharina von der Wense, Anne Lauscher Published: 2026-03-23 arXiv: 2603.22260
Summary
Hundreds of millions of people rely on large language models (LLMs) for education, work, and even healthcare. Yet these models are known to reproduce and amplify social biases present in their training data. Moreover, text-based interfaces remain a barrier for many users — for example, those with limited literacy, motor impairments, or mobile-only devices. Voice interaction promises to expand accessibility, but unlike text, speech carries identity cues that users cannot easily mask, raising concerns about whether accessibility gains may come at the cost of equitable treatment.
Key Findings
- Audio-enabled LLMs exhibit systematic gender discrimination, shifting responses toward gender-stereotyped adjectives and occupations solely on the basis of speaker voice.
- The observed voice-based bias amplifies the discrimination found in text-based interaction.
- Voice interfaces are not merely an extension of text models to a new modality; they introduce distinct bias mechanisms tied to paralinguistic cues.
- This creates a tension: improving accessibility (via speech) may simultaneously expose users to greater inequitable treatment.
Original Abstract (excerpt)
> Hundreds of millions of people rely on large language models (LLMs) for education, work, and even healthcare. Yet these models are known to reproduce and amplify social biases present in their training data. Moreover, text-based interfaces remain a barrier for many, for example, users with limited literacy, motor impairments, or mobile-only devices. Voice interaction promises to expand accessibility, but unlike text, speech carries identity cues that users cannot easily mask, raising concerns about whether accessibility gains may come at the cost of equitable treatment. Here we show that audio-enabled LLMs exhibit systematic gender discrimination, shifting responses toward gender-stereotyped adjectives and occupations solely on the basis of speaker voice...
---
*Auto-collected on 2026-03-25.*