Exploring Gender Disparities in Automatic Speech Recognition Technology

2025-02-25

AI Summary

This study investigates gender fairness in automatic speech recognition (ASR), moving beyond demographic parity to systematically analyze how training data gender composition and acoustic characteristics—particularly pitch variability—affect model bias. Using the LibriSpeech corpus and the Whisper-small model, we integrate acoustic feature analysis with multidimensional fairness quantification. Our findings reveal: (1) optimal gender fairness does not occur at a 50–50 training data split; instead, a specific, non-uniform gender ratio maximizes fairness; and (2) pitch variability is a critical acoustic driver of gender bias—an empirically validated insight not previously established. We characterize the nonlinear relationship between training data gender composition and ASR fairness, providing reproducible guidelines for data balancing and actionable strategies for bias mitigation. This work establishes a methodological foundation for developing equitable ASR systems.

Technology Category

• Machine Learning: Ethics, Bias, and Fairness
• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy
• Computer Vision: Bias, Fairness & Privacy

Application Category

• Social Networks and Social Media: Fairness and bias in social network and social media analysis
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

This study investigates factors influencing Automatic Speech Recognition (ASR) systems' fairness and performance across genders, beyond the conventional examination of demographics. Using the LibriSpeech dataset and the Whisper small model, we analyze how performance varies across different gender representations in training data. Our findings suggest a complex interplay between the gender ratio in training data and ASR performance. Optimal fairness occurs at specific gender distributions rather than a simple 50-50 split. Furthermore, our findings suggest that factors like pitch variability can significantly affect ASR accuracy. This research contributes to a deeper understanding of biases in ASR systems, highlighting the importance of carefully curated training data in mitigating gender bias.

Problem

• Analyzing ASR system fairness across genders
• Investigating gender ratio impact on ASR performance
• Exploring pitch variability effects on ASR accuracy

Innovation

• Analyzes ASR performance across genders
• Uses LibriSpeech and Whisper model
• Examines training data gender ratio impact