Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Robust Generative Audio Quality Assessment: Disentangling Quality from Spurious Correlations

Kuan-Tang Huang, Chien-Chun Wang, Cheng-Yeh Yang, Hung-Shin Lee, Hsin-Min Wang, Berlin Chen

Abstract

The rapid proliferation of AI-Generated Content (AIGC) has necessitated robust metrics for perceptual quality assessment. However, automatic Mean Opinion Score (MOS) prediction models are often compromised by data scarcity, predisposing them to learn spurious correlations-- such as dataset-specific acoustic signatures-- rather than generalized quality features. To address this, we leverage domain adversarial training (DAT) to disentangle true quality perception from these nuisance factors. Unlike prior works that rely on static domain priors, we systematically investigate domain definition strategies ranging from explicit metadata-driven labels to implicit data-driven clusters. Our findings reveal that there is no "one-size-fits-all" domain definition; instead, the optimal strategy is highly dependent on the specific MOS aspect being evaluated. Experimental results demonstrate that our aspect-specific domain strategy effectively mitigates acoustic biases, significantly improving correlation with human ratings and achieving superior generalization on unseen generative scenarios.

Robust Generative Audio Quality Assessment: Disentangling Quality from Spurious Correlations

Abstract

The rapid proliferation of AI-Generated Content (AIGC) has necessitated robust metrics for perceptual quality assessment. However, automatic Mean Opinion Score (MOS) prediction models are often compromised by data scarcity, predisposing them to learn spurious correlations-- such as dataset-specific acoustic signatures-- rather than generalized quality features. To address this, we leverage domain adversarial training (DAT) to disentangle true quality perception from these nuisance factors. Unlike prior works that rely on static domain priors, we systematically investigate domain definition strategies ranging from explicit metadata-driven labels to implicit data-driven clusters. Our findings reveal that there is no "one-size-fits-all" domain definition; instead, the optimal strategy is highly dependent on the specific MOS aspect being evaluated. Experimental results demonstrate that our aspect-specific domain strategy effectively mitigates acoustic biases, significantly improving correlation with human ratings and achieving superior generalization on unseen generative scenarios.
Paper Structure (13 sections, 1 equation, 5 figures, 2 tables)

This paper contains 13 sections, 1 equation, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Proposed Method
  3. Model Architecture
  4. Domain Definition Strategies
  5. Experimental Setup
  6. Dataset
  7. Training Setup
  8. Evaluation Metrics
  9. Results and Discussion
  10. Main results
  11. Latent Space Analysis
  12. Impact of Domain Granularity and Grouping Strategy
  13. Conclusion and Future Work

Figures (5)

  • Figure 1: The proposed model architecture with DAT.
  • Figure 2: Performance comparison on Audiobox-Aesthetics across MSE and SRCC. The results are reported for four aspects: PQ, PC, CE, and CU.
  • Figure 3: Visualization of the $\mathbf{h}$ on the development set using UMAP. The top row is colored by source domain labels, and the bottom by PC scores.
  • Figure 4: 3D "Quality Terrain" generated by combining 2D UMAP projections of encoder features $\mathbf{h}$ with the predicted MOS as the z-axis for (a) the baseline and (b) our proposed DAT strategy.
  • Figure 5: Ablation study on domain granularity $K$ for the PQ dimension. The top panel shows the absolute improvement in SRCC ($\Delta$ SRCC), and the bottom panel shows the improvement in MSE ($\Delta$ MSE) relative to the baseline. The star ($\star$) denotes the optimal configuration at $K=8$, which yields the most balanced gains across both metrics.