Current ASR fairness benchmarks using single heterogeneous speaker groups can misidentify mistreated groups, so evaluations should use fine-grained intersectional analysis of available demographic metadata.
However, there are several corpora specifically designed for bias/fairness evaluation of ASR systems whose multitudinous metadata cat- egories permit finer-grained ASR evaluation
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Responsible Benchmarking of Fairness for Automatic Speech Recognition
Current ASR fairness benchmarks using single heterogeneous speaker groups can misidentify mistreated groups, so evaluations should use fine-grained intersectional analysis of available demographic metadata.