On the Robust Approximation of ASR Metrics

Abdul Waheed Hanin Atwany Rita Singh Bhiksha Raj · 2025

Read On the Robust Approximation of ASR Metrics by Abdul Waheed Hanin Atwany Rita Singh Bhiksha Raj (2025) 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-2502-12408.txt.

Description

Abdul Waheed and co-authors guess speech scores with no true scripts in a 2025 paper. Speech models are graded by word and letter error rates against true text. True labels are scarce for wild, real world speech, and costly to make.

Their method joins speech and text codes in one shared space. A strong proxy model gives proxy scores for each clip. A regression model then learns to map those traits to true error rates. Tests on forty plus models and fourteen data sets landed near true scores. Label free grades can show how models fare far past the bench mark sets.

Catalog record

CREATORAbdul Waheed Hanin Atwany Rita Singh Bhiksha Raj
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE79 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-12408

Browse the full library · About the Tower of Babel