Does KAD fix FAD's Gaussian and sample-size problems, and how large must N be?

Problem: choose a distribution-distance metric for small evaluation sets. Conditions: audio generation experiments with limited generated samples. Method (Chung et al., 2025): FAD assumes Gaussian embeddings, which often fail (the paper shows non-Gaussian VGGish embeddings), and its finite-sample bias scales as O(1/N), so FAD decreases as N grows and can be manipulated by changing N. Kernel Audio Distance uses Maximum Mean Discrepancy (distribution-free, unbiased) and the kadtk toolkit. Measured by the authors: on Clotho 2 with N from 100 to 3,839 the FAD deviation roughly halves each time sample size doubles, while KAD stays near its asymptotic value; on DCASE 2023 Task 7 Foley systems, KAD reaches a Spearman correlation of up to -0.93 with human ratings against -0.80 for FAD's best case. Failure modes: KAD has larger variance at small N; music-trained embeddings (MERT, CLAP-laion-music) correlated less well than others on Foley. Limitations: authors' own comparison, Foley sound not music; negative sign is because lower distance should mean better. Version/date: arXiv v2 2025-03-09. Curator qualifications: All cited documentation excerpts located on the live pages by curator GET; no video sources. Limitations: Foley-sound evidence only; no music listening test in this paper. Evidence status: statements located in the cited sources by a correlated desk review (Claude); not executed; not independently reviewed.

{
  "citations": [
    {
      "url": "https://arxiv.org/pdf/2502.15602",
      "note": "Abstract; 3 Sample size bias; 3 Sample size bias; 4 Human alignment"
    }
  ]
}