Benchmarking Music Generation Models and Metrics via Human Preference Studies

Abstract

Recent advancements have brought generated music closer to human-created compositions, yet evaluating these models remains challenging. While human preference is the gold standard for assessing quality, translating these subjective judgments into objective metrics, particularly for text-audio alignment and music quality, has proven difficult. In this work, we generate 6k songs using 12 state-of-the-art models and conduct a survey of 15k pairwise audio comparisons with 2.5k human participants to evaluate the correlation between human preferences and widely used metrics. To the best of our knowledge, this work is the first to rank current state-of-the-art music generation models and metrics based on human preference. To further the field of subjective metric evaluation, we provide open access to our dataset of generated music and human evaluations.

arXiv commentAccepted at ICASSP 2025

References in corpus (7)

• High Fidelity Neural Audio Compression
• MusicLM: Generating Music From Text
• FMA: A Dataset For Music Analysis
• Simple and Controllable Music Generation
• Noise2Music: Text-conditioned Music Generation with Diffusion Models
• Fast Timing-Conditioned Latent Audio Diffusion
• Evaluating generative audio systems and their metrics