Musicprefs: Pairwise Human Preferences for Text-to-Music Systems

A collection of pairwise human preferences for music generated by text-to-music systems. The dataset is hosted on Huggingface Datasets by the author 'i-need-sleep' and was last updated on 2026-01-27. It is intended for research into evaluating and improving generative music models.

Use Cases

• Training or fine-tuning reward models for music generation based on human preference data.
• Benchmarking the output quality of different text-to-music systems.
• Analyzing patterns in human preferences for AI-generated music.
• Conducting research on alignment techniques for generative audio models.

Strengths

• Contains human preference data, which is a direct signal for model alignment.
• Specifically designed for evaluating text-to-music systems, providing a focused use case.
• Last updated on 2026-01-27, suggesting potential recent curation.

Limitations

• Row count, file formats, and dataset size are unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• The description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

huggingface

Collection Method

Human annotation of pairwise comparisons between AI-generated music samples.

Freshness

Last updated 2026-01-27 10:00:50.