Reproduce “Route-DMD: Distribution Matching Distillation with Task-Routed Mechanism for Speech Foundation Models” 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
Yanru Huo, Zhikang Niu, Wenming Tu, Junxi Liu, Tian Tan, Shan Yang, Yipeng Kang, Ziyang Ma and 2 more
Route-DMD accelerates a speech model for generation, editing, and processing by distilling most tasks with Decoupled Distribution Matching Distillation and routing separation examples to regression against paired targets. A consistency-distillation stage initializes the student, and adaptive projected guidance replaces classifier-free guidance in teacher rollouts. Its four-step student achieves a 4.5× speedup over the 32-step teacher, while Libri2Mix word error rate worsens.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.