Reproduce “TTSonic: A Unified Framework for Composable and Controllable Speech Generation” 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
Zhedong Zhang, Kaiwen Zhou, Junyan Wu, Shengqiang Li, Xuran Chi, Xingui Zeng, Rui Yan, Xu Chen
TTSonic combines a tokenizer, language model, and flow-matching renderer to support speech-generation tasks including voice conversion. Its tokenizer supplies speech tokens, while the renderer conditions them on a speaker embedding; conversion uses source speech tokens with a different target speaker’s condition, despite no paired conversion training. On Chinese Seed VC, WER is 1.85% versus X-VC’s 1.99%; English WER trails.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.