Reproduce “Edit Who Speaks, Control How They Speak: Global Timbre Editing and Local Instruction Control for TTS” 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

Junchuan Zhao, Chenglin Xu, Wei Zeng, Haoyang Li, Yiwen Guo, Ye Wang

EDICT edits a voice reference and supports segment-by-segment delivery instructions in text-to-speech. It encodes global edits across nine vocal attributes, then refreshes a frozen model’s conditioning at instruction changes while retaining selected acoustic context. On IntraTTS-Bench, transitions scored 82.00% versus 64.94% with truncated-cache reuse, though recognition errors were higher; results are benchmark-specific.

Summary by alphaXiv.

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