Gender Bias and Sonic Gender Performance in AI-Generated Music: Representation, Expression, and Perception

Abstract

Gender bias in generative AI has typically been studied in text and image domains, but music-generating AI poses a different challenge: in music, gender can be expressed through sound, style, genre, and multimodal presentation as well as explicit representation. Through a case study of Suno, this study examines gender bias not only in vocal-gender representation, but also in generated musical expression and listener perception. First, generated vocal music shows a more extreme male-dominant pattern across genres than real-world music metadata. To further examine how gender is musically realized, we generated non-vocal instrumental music using gendered prompts. Gendered bias appeared through distributed stylistic cues, including genre descriptions, title words, cover images, audio features, and symbolic MIDI features. In genre-controlled generations, masculine-prompted outputs were often farther from neutral outputs than feminine-prompted outputs were, suggesting a stronger stylistic distinction for masculinity. Listeners’ gendered impressions were shaped largely by mood and culturally familiar genre associations. Together, these findings suggest that gender bias in music-generating AI appears across multiple layers, involving how music and gender are repeatedly connected, expressed, and perceived throughout generation and listening.