Toward Fair Speech Technologies: A Comprehensive Survey of Bias and Fairness in Speech AI
Yi-Cheng Lin, Yun-Shao Tsai, Kuan-Yu Chen, Hsiao-Ying Huang, Huang-Cheng Chou, and Hung-yi Lee · 2026
Read Toward Fair Speech Technologies: A Comprehensive Survey of Bias and Fairness in Speech AI by Yi-Cheng Lin, Yun-Shao Tsai, Kuan-Yu Chen, Hsiao-Ying Huang, Huang-Cheng Chou, and Hung-yi Lee (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-01597.txt.
Introduction
Yi-Cheng Lin and eight co-authors survey bias and fairness across speech technology, pulling over four hundred studies into one frame. Speech tools now gate high-stakes services, yet fairness work has been split by task and field, so patterns shared across the domain stay hidden.
The survey sets out seven fairness definitions tuned to speech and tracks the field's growth through three paradigms: robustness, representation, and governance. Metrics are sorted into six families, each tied back to the definition it serves. Bias is traced along the whole pipeline, surfacing speech-specific traps such as recording channel acting as a stand-in for demographic traits, and shaky human labels for emotion. Fixes are organized by the stage where they intervene, from data to deployment, and matched to the sources they treat. Open challenges close the map.
Catalog record
| CREATOR | Yi-Cheng Lin, Yun-Shao Tsai, Kuan-Yu Chen, Hsiao-Ying Huang, Huang-Cheng Chou, and Hung-yi Lee |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 144 KB |
| LICENSE | http://creativecommons.org/licenses/by-sa/4.0/ |
| RIGHTS | CC BY SA |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.01597