Reproduce “Learnable Classifier-Free Guidance Null Embeddings for Enhanced Controllable Speech Synthesis” 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
Biel Tura Vecino, Yoach Lacombe, Julian Weber, Zbigniew Łatka, Haitong Zhang, Logan Hart, Eren Gölge
The authors introduce learnable null embeddings for text and speaker conditions in a text-to-speech system, replacing fixed zero vectors used for classifier-free guidance. In their comparison, the learned-null model had higher ECAPA2 speaker similarity (SECS 0.817 versus 0.755) and greater pitch variation. CER was 0.9 versus 1.2, a difference reported as not statistically meaningful; evidence is limited to their custom evaluation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.