A Paradigm for Interpreting Metrics and Identifying Critical Errors in Automatic Speech Recognition

Thibault Bañeras-Roux 1 , Mickael Rouvier 2 , Jane Wottawa 3 , Richard Dufour 1 · 2026

Read A Paradigm for Interpreting Metrics and Identifying Critical Errors in Automatic Speech Recognition by Thibault Bañeras-Roux 1 , Mickael Rouvier 2 , Jane Wottawa 3 , Richard Dufour 1 (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-2605-03671.txt.

Description

Thibault Bañeras-Roux, Mickael Rouvier, Jane Wottawa, and Richard Dufour propose a new way to judge automatic speech recognition errors by how humans perceive them. Word and character error rates are easy to interpret but correlate poorly with human judgment, while embedding-based metrics match perception better yet produce scores that are hard to read.

Their paradigm folds a chosen perceptual metric into a Minimum Edit Distance measure, an error-rate equivalent whose units stay interpretable. Errors are aligned with their perceived severity, which also allows an original study of which transcription mistakes humans find most serious. The approach aims to keep the clarity of classic error rates while grounding the numbers in human perception.

Catalog record

CREATORThibault Bañeras-Roux 1 , Mickael Rouvier 2 , Jane Wottawa 3 , Richard Dufour 1
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE20 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.03671

Browse the full library · About the Tower of Babel