Reproduce “Tacit-TTS: From Autoregressive Decoding to Masked Prediction for Efficient Transcript-Free 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
Jian Chen, You Zhang, Mark Vinton
Tacit-TTS uses masked prediction to generate speech-content tokens in parallel instead of autoregressive text-to-semantic generation. It conditions on reference audio, estimates output length from speaking pace and target text, and renders speech through a distilled IndexTTS2 pipeline. On an A100, generation-stage latency fell from 6,149 to 509 milliseconds, or 12.1×, for 32 utterances; speaker similarity remained close, but WER increased.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.