A Large-Scale Benchmark Dataset for Aesthetic Evaluation of Complete Songs
📖 Overview
SongEval is the first open-source, large-scale benchmark dataset designed for aesthetic evaluation of complete songs. It provides over 2,399 songs (~140 hours) annotated by 16 expert raters across five perceptual dimensions. The dataset enables research in evaluating and improving music generation systems from a human aesthetic perspective.
🌟 Features… See the full description on the dataset page: https://huggingface.co/datasets/ASLP-lab/SongEval.