The T12 System for AudioMOS Challenge 2025: Audio Aesthetics Score Prediction System Using KAN- and VERSA-based Models

2025-12-05

AI Summary

This study addresses the audio aesthetic scoring (AES) task and its four sub-dimension prediction subtasks. We propose a multimodal regression framework integrating the Kolmogorov–Arnold Network (KAN) with the VERSA architecture. Methodologically, we replace conventional MLPs with group-wise rational KANs and employ a pseudo-labeling strategy to augment training data; additionally, we apply XGBoost to fuse heterogeneous audio metrics and model predictions for robust ensemble scoring. To our knowledge, this is the first systematic application of KANs to audio aesthetic modeling. By synergizing multi-model collaboration and metric-driven fusion, our approach significantly improves cross-dimensional consistency. Our method achieves the highest Spearman correlation on all evaluation levels—utterance-level (3 dimensions), system-level (2 dimensions), and overall average metrics—securing first place in the challenge.

Technology Category

• Machine Learning: Multimodal Learning
• Cognitive Modeling & Cognitive Systems: Affective Computing
• Computer Vision: Multi-modal Vision

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Metrics for user behavior and evaluating success

Abstract

We propose an audio aesthetics score (AES) prediction system by CyberAgent (AESCA) for AudioMOS Challenge 2025 (AMC25) Track 2. The AESCA comprises a Kolmogorov--Arnold Network (KAN)-based audiobox aesthetics and a predictor from the metric scores using the VERSA toolkit. In the KAN-based predictor, we replaced each multi-layer perceptron layer in the baseline model with a group-rational KAN and trained the model with labeled and pseudo-labeled audio samples. The VERSA-based predictor was designed as a regression model using extreme gradient boosting, incorporating outputs from existing metrics. Both the KAN- and VERSA-based models predicted the AES, including the four evaluation axes. The final AES values were calculated using an ensemble model that combined four KAN-based models and a VERSA-based model. Our proposed T12 system yielded the best correlations among the submitted systems, in three axes at the utterance level, two axes at the system level, and the overall average.

Problem

• Predict audio aesthetics scores using KAN and VERSA models
• Enhance scoring accuracy via ensemble of neural and regression predictors
• Evaluate performance across multiple axes and system levels

Innovation

• KAN-based model with group-rational layers
• VERSA-based regression using gradient boosting
• Ensemble combining KAN and VERSA predictors