Reproduce “VIPBench: A Human-Aligned Benchmark for Voice Identity Perception in the Age of Voice Cloning” 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
Xudong Tang, Matthew Groh
VIPBench evaluates speaker models against listeners’ perceived voice identity across real recordings, clones, and morphs. It contains 124,876 judgments on 9,800 pairs; listeners answered “different” in 41% of responses to same-speaker real-and-clone pairs. Resemblyzer best predicts perceived sameness (Pearson r = 0.656; 61% of explainable variance). Real-pair calibration transfers less well to the evaluated synthetic recordings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.