Emotional Misalignment in Contemporary Music: Systematic Patterns of Audio-Lyrical Contrast Across Genres
Year: 2025
Type: dissertation
Abstract: This thesis investigates emotional alignment between audio features and lyrical sentiment in contemporary popular music, addressing a gap in music emotion research through a novel dual-modality approach. The study analyzed 1,080 songs across 12 genres using K-means clustering of Spotify audio features and GPT-4 lyrical emotion classification, identifying three emotional clusters: Aggressive/Intense, Sad/Calm, and Happy/Upbeat. Key findings reveal that emotional misalignment is the norm rather than the exception: only 46% of songs showed audio-lyrical alignment, while 54% exhibited systematic emotional contrast. Statistical analysis confirmed significant relationships between modalities (χ2 = 99.45, p < 0.001) but slight practical agreement (κ = 0.164), indicating systematic yet divergent patterns. Genre analysis revealed distinct strategies: hip-hop featured angry lyrics with happy audio (55% mismatches), jazz combined happy lyrics with sad audio (92% mismatches), and metal paired sad lyrics with angry audio (66% mismatches). The research demonstrates that emotional misalignment represents intentional artistic choices creating irony, amplification, or accessibility rather than anomalies. These findings have significant implications for music recommendation systems, therapeutic applications, and AI-generated music, suggesting need for nuanced approaches accounting for multi-modal complexity.
Author: Attia, Mohamed Taha Ben
Language: English
Topic: Music and Audio Processing
Subfield: Signal Processing
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7113798611