Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

The ICASSP 2026 Automatic Song Aesthetics Evaluation Challenge

Guobin Ma, Yuxuan Xia, Jixun Yao, Huixin Xue, Hexin Liu, Shuai Wang, Hao Liu, Lei Xie

TL;DR

Addresses the problem of evaluating subjective song aesthetics for AI-generated music without reference signals. Proposes a two-track benchmark with a hierarchical scoring framework that yields a per-utterance, per-system composite $S_{set} = \frac{1}{4} \left( \mathrm{TTA}_{u} + \sum_{m \in \mathcal{M}} \frac{m_{u} + m_{s}}{2} \right)$ and track-specific aggregations for ranking: $ \text{Score}_{Track1} = 0.2 \times S_{\text{set1}} + 0.8 \times S_{\text{set2}}$ and $ \text{Score}_{Track2} = \frac{1}{5} \sum_{d=1}^{5} S_d$. The results show that top systems surpass the baseline by leveraging multi-representation fusion and structure-aware modeling, with strong generalization on both Regular and Hard splits but modest Top-Tier Accuracy. The work provides a standardized benchmark and evaluation methodology that can guide optimization and RL-based approaches in modern music generation.

Abstract

This paper summarizes the ICASSP 2026 Automatic Song Aesthetics Evaluation (ASAE) Challenge, which focuses on predicting the subjective aesthetic scores of AI-generated songs. The challenge consists of two tracks: Track 1 targets the prediction of the overall musicality score, while Track 2 focuses on predicting five fine-grained aesthetic scores. The challenge attracted strong interest from the research community and received numerous submissions from both academia and industry. Top-performing systems significantly surpassed the official baseline, demonstrating substantial progress in aligning objective metrics with human aesthetic preferences. The outcomes establish a standardized benchmark and advance human-aligned evaluation methodologies for modern music generation systems.

The ICASSP 2026 Automatic Song Aesthetics Evaluation Challenge

TL;DR

Addresses the problem of evaluating subjective song aesthetics for AI-generated music without reference signals. Proposes a two-track benchmark with a hierarchical scoring framework that yields a per-utterance, per-system composite and track-specific aggregations for ranking: and . The results show that top systems surpass the baseline by leveraging multi-representation fusion and structure-aware modeling, with strong generalization on both Regular and Hard splits but modest Top-Tier Accuracy. The work provides a standardized benchmark and evaluation methodology that can guide optimization and RL-based approaches in modern music generation.

Abstract

This paper summarizes the ICASSP 2026 Automatic Song Aesthetics Evaluation (ASAE) Challenge, which focuses on predicting the subjective aesthetic scores of AI-generated songs. The challenge consists of two tracks: Track 1 targets the prediction of the overall musicality score, while Track 2 focuses on predicting five fine-grained aesthetic scores. The challenge attracted strong interest from the research community and received numerous submissions from both academia and industry. Top-performing systems significantly surpassed the official baseline, demonstrating substantial progress in aligning objective metrics with human aesthetic preferences. The outcomes establish a standardized benchmark and advance human-aligned evaluation methodologies for modern music generation systems.
Paper Structure (6 sections, 3 equations, 1 figure)

This paper contains 6 sections, 3 equations, 1 figure.

Table of Contents

  1. Introduction
  2. Dataset
  3. Challenge Metrics
  4. Track 1: Overall Musicality Prediction
  5. Track 2: Fine-grained Dimension Prediction
  6. Results and Conclusion

Figures (1)

  • Figure 1: Overview of the challenge results. The visualizations depict the performance distribution of participating teams in Track 1 (Overall Musicality) and Track 2 (Fine-grained Dimensions) compared to the baseline system.