Reproduce “Quantifying Source Speaker Leakage in One-to-One Voice Conversion” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Scott Wellington, Xuechen Liu, Junichi Yamagishi

Researchers quantify the degree of source speaker information leakage in one-to-one voice conversion systems through a novel measurement approach using Earth Mover's Distance, revealing varying levels of privacy risk based on speaker characteristics like gender, accent, and recording environment.

Key insights

  • Gender mismatches between source and target speakers result in the highest information leakage
  • Accent mismatches show varying levels of leakage, with Scottish accents leading to higher leakage than Midlands/Welsh
  • Information leakage decreases when source and target speakers are closer in age
  • Recording environment differences contribute to detectable information leakage

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.