Reproduce “Zero-VC: Zero-Lookahead Streaming Voice Conversion via Speaker Anonymization” 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
Yudong Li, Zihao Fang, Junwen Qiu, Ruihai Jing, Ruixiang Hang, Yingda Shen, Zhizheng Wu
A streaming zero-shot voice conversion system named Zero-VC is introduced, achieving a minimal algorithmic latency of 20 ms. It leverages Speaker Anonymization (SA) to effectively disentangle speaker identity while preserving linguistic and prosodic content, enabling high-quality voice conversion for unseen speakers in real-time applications.
Key insights
- Speaker Anonymization (SA) provides a robust content representation for streaming VC, effectively disentangling speaker identity with minimal timbre leakage while retaining rich prosodic and linguistic utility.
- SA-perturbed features significantly reduce the generative decoder's reliance on future contextual information, enabling a truly zero-lookahead (20 ms algorithmic latency) architecture.
- Explicitly optimizing the trade-off between timbre leakage and utility preservation through SA allows for superior zero-shot VC quality under stringent real-time constraints.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.