Beyond Genre: Diagnosing Bias in Music Embeddings Using Concept Activation Vectors

2025/09/29 by Roman B. Gebhardt, Gebhardt, Roman B., Arne Kuhle +3
Computer Science · Social Sciences · #Advanced Text Analysis Techniques #Audio and Speech Processing (eess.AS) #Computational and Text Analysis Methods #FOS: Computer and information sciences #FOS: Electrical engineering #Music and Audio Processing #Sound (cs.SD) #electronic engineering #information engineering

paper · pdf · doi:10.48550/arxiv.2509.24482

openalex publication_date 2025/09/29 · openalex created_date 2025/10/19 · openalex updated_date 2026/07/28

Abstract

Music representation models are widely used for tasks such as tagging, retrieval, and music understanding. Yet, their potential to encode cultural bias remains underexplored. In this paper, we apply Concept Activation Vectors (CAVs) to investigate whether non-musical singer attributes - such as gender and language - influence genre representations in unintended ways. We analyze four state-of-the-art models (MERT, Whisper, MuQ, MuQ-MuLan) using the STraDa dataset, carefully balancing training sets to control for genre confounds. Our results reveal significant model-specific biases, aligning with disparities reported in MIR and music sociology. Furthermore, we propose a post-hoc debiasing strategy using concept vector manipulation, demonstrating its effectiveness in mitigating these biases. These findings highlight the need for bias-aware model design and show that conceptualized interpretability methods offer practical tools for diagnosing and mitigating representational bias in MIR.

Citations

• Post-Hoc Concept Disentanglement: From Correlated to Isolated Concept Representations
• Understanding and Mitigating Gender Bias in LLMs via Interpretable Neuron Editing
• MuQ: Self-Supervised Music Representation Learning with Mel Residual Vector Quantization
• MuChoMusic: Evaluating Music Understanding in Multimodal Audio-Language Models
• Let Network Decide What to Learn: Symbolic Music Understanding Model Based on Large-scale Adversarial Pre-training
• STraDa: A Singer Traits Dataset
• LyricWhiz: Robust Multilingual Zero-shot Lyrics Transcription by Whispering to ChatGPT
• MERT: Acoustic Music Understanding Model with Large-Scale Self-supervised Training
• Learning Disentangled Semantic Spaces of Explanations via Invertible Neural Networks
• Robust Speech Recognition via Large-Scale Weak Supervision
• Concept-Based Techniques for "Musicologist-friendly" Explanations in a Deep Music Classifier
• Learning Unsupervised Hierarchies of Audio Concepts
• Human-Centered Concept Explanations for Neural Networks
• Assessing Algorithmic Biases for Musical Version Identification
• Investigating Bias in Image Classification using Model Explanations
• Exploring Artist Gender Bias in Music Recommendation
• Concept Bottleneck Models
• Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word\n Embeddings

Related

• Tuning Into Bias: A Computational Study of Gender Bias in Song Lyrics
• Evaluation of pretrained language models on music understanding
• Music for All: Representational Bias and Cross-Cultural Adaptability of Music Generation Models
• Quantifying Gender Bias in Consumer Culture
• A Study on the Data Distribution Gap in Music Emotion Recognition