English

Discrete optimal transport is a strong audio adversarial attack

Audio and Speech Processing 2026-02-20 v2 Artificial Intelligence

Abstract

In this paper, we introduce the discrete optimal transport voice conversion (kkDOT-VC) method. Comparison with kkNN-VC, SinkVC, and Gaussian optimal transport (MKL) demonstrates stronger domain adaptation abilities of our method. We use the probabilistic nature of optimal transport (OT) and show that kkDOT-VC is an effective black-box adversarial attack against modern audio anti-spoofing countermeasures (CMs). Our attack operates as a post-processing, distribution-alignment step: frame-level {WavLM} embeddings of generated speech are aligned to an unpaired bona fide pool via entropic OT and a top-kk barycentric projection, then decoded with a neural vocoder. Ablation analysis indicates that distribution-level alignment is a powerful and stable attack for deployed CMs.

Cite

@article{arxiv.2509.14959,
  title  = {Discrete optimal transport is a strong audio adversarial attack},
  author = {Anton Selitskiy and Akib Shahriyar and Jishnuraj Prakasan},
  journal= {arXiv preprint arXiv:2509.14959},
  year   = {2026}
}
R2 v1 2026-07-01T05:43:53.437Z