FairASR pretrains a Conformer with InfoNCE plus a gradient-reversed supervised contrastive loss over demographic labels, reducing demographic WER gaps on FairSpeech with small overall WER cost.
Our method explicitly utilizes demographic labels during training, ensuring that learned representations remain robust, discrimina- tive, and fair across diverse populations
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
eess.AS 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
FairASR: Fair Audio Contrastive Learning for Automatic Speech Recognition
FairASR pretrains a Conformer with InfoNCE plus a gradient-reversed supervised contrastive loss over demographic labels, reducing demographic WER gaps on FairSpeech with small overall WER cost.