Reproduce “InstCharVoice: Grounding Natural-Language Instructions for Character-Level Control in Text-to-Speech” 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
Sihang Nie, Xueru Li, Xiaofen Xing, Deyi Tuo, Cheng-Bin Jin, Jingyuan Xing, Jinxin Ji
InstCharVoice connects natural-language instructions to character-level acoustic plans for Chinese speech. It identifies instruction-relevant characters, predicts duration, boundary, energy, pitch, and tone, then generates speech character by character. On WordVoice-5A-zh, its keyword-level boundary relaxed error rate is 14.04%, versus 24.27% for CosyVoice3. This measures localized matching, not perfect realization or generalization to all instructions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.