A Contrastive Learning Approach to Mitigate Bias in Speech Models

TL;DR

This paper introduces a contrastive learning method to reduce bias in speech models by improving subgroup representations, leading to fairer and more balanced performance across diverse populations.

Contribution

It is the first to apply contrastive learning specifically for mitigating bias in speech models at the subgroup level.

Findings

Improves internal subgroup representations

Reduces model bias across subgroups

Enhances performance on spoken language understanding tasks

Abstract

Speech models may be affected by performance imbalance in different population subgroups, raising concerns about fair treatment across these groups. Prior attempts to mitigate unfairness either focus on user-defined subgroups, potentially overlooking other affected subgroups, or do not explicitly improve the internal representation at the subgroup level. This paper proposes the first adoption of contrastive learning to mitigate speech model bias in underperforming subgroups. We employ a three-level learning technique that guides the model in focusing on different scopes for the contrastive loss, i.e., task, subgroup, and the errors within subgroups. The experiments on two spoken language understanding datasets and two languages demonstrate that our approach improves internal subgroup representations, thus reducing model bias and enhancing performance.