Robust Generative Audio Quality Assessment: Disentangling Quality from Spurious Correlations
This work addresses the need for robust perceptual quality assessment in AI-generated audio, offering an incremental improvement by optimizing domain strategies for specific MOS aspects to mitigate biases.
The paper tackles the problem of automatic Mean Opinion Score (MOS) prediction for AI-generated audio, which is often biased by spurious correlations like dataset-specific acoustic signatures, by using domain adversarial training to disentangle true quality perception from these factors. The result is a method that significantly improves correlation with human ratings and achieves superior generalization on unseen generative scenarios.
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.