On the social bias of speech self-supervised models

TL;DR

This paper investigates social bias in self-supervised speech models, revealing how architecture and training influence bias propagation and demonstrating that regularization and model compression can effectively reduce such biases.

Contribution

It is the first comprehensive analysis of social bias in SSL speech models, exploring factors affecting bias and proposing debiasing methods through regularization and model compression.

Findings

Bias is present in prevalent SSL models affecting marginalized groups.

Model architecture and training methods influence bias levels.

Regularization and compression techniques can mitigate social bias.

Abstract

Self-supervised learning (SSL) speech models have achieved remarkable performance in various tasks, yet the biased outcomes, especially affecting marginalized groups, raise significant concerns. Social bias refers to the phenomenon where algorithms potentially amplify disparate properties between social groups present in the data used for training. Bias in SSL models can perpetuate injustice by automating discriminatory patterns and reinforcing inequitable systems. This work reveals that prevalent SSL models inadvertently acquire biased associations. We probe how various factors, such as model architecture, size, and training methodologies, influence the propagation of social bias within these models. Finally, we explore the efficacy of debiasing SSL models through regularization techniques, specifically via model compression. Our findings reveal that employing techniques such as row-pruning and training wider, shallower models can effectively mitigate social bias within SSL model.