What does the Seed-VC paper claim about zero-shot voice conversion (timbre leakage, similarity, intelligibility), and how big was the evaluation?

Problem. Voice conversion must keep what was said and how, while replacing who says it; content features often leak source timbre, and removing leakage can raise word error rate. Conditions. Zero-shot: source and target speakers unseen in training; reference audio of a few seconds. Method. Seed-VC uses a diffusion transformer with in-context conditioning on reference audio and an external timbre shifter during training so content features do not carry the source timbre. Measured result (author-reported). On 100 LibriTTS test-clean source utterances and 8 target-speaker references (seed-tts-eval), Seed-VC reached speaker similarity (SECS) 0.8676 and WER 11.99% against ground-truth WER 8.02% and OpenVoice SECS 0.7547; the paper also reports lower CER than baselines. Failure modes. The authors note OpenVoice scored slightly higher on some DNSMOS quality scores, interpreted as a trade-off with timbre fidelity. Limits. Small eval (8 targets), English-centric, objective metrics only, no listening test reported in the text I read; single author, arXiv preprint. Version/date. arXiv v1, 15 Nov 2024. Curator qualifications: All cited documentation excerpts located on the live pages by curator GET; no video sources. Limitations: Author-reported on a small objective evaluation; ground-truth WER 8.02 read from the extracted table, verify against the PDF. 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/2411.09943",
      "note": "Abstract; Sec 4.1; Sec 4.2; Sec 4.2"
    }
  ]
}