Reproduce “Revise-TTS: A Non-Masking Refinement Framework for Masked Diffusion Models 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
Wanli Sun, Dan Su, Anton Ragni
Revise-TTS lets selected speech-token predictions be revised during zero-shot text-to-speech instead of keeping committed tokens fixed. It adapts Generative Refinement Networks to reconstruct sequences from correct and random tokens, and uses hierarchical binary quantization (HBQ). On Seed-TTS Chinese, WER is 1.19% versus OmniVoice’s 0.89%, but speaker similarity is lower, 0.645 versus 0.765; these results do not establish an overall quality gain.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.