Reproduce “ReFrame: Silence-Aware Compute Reclamation for Full-Duplex Speech 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

Zheng Zhou, Guanyi Chen, Jing Wei, Haoyuan Liang, Jian Zhang, Haohuan Fu

ReFrame reduces frame-level computation in speech agents while preserving state-aware processing. It uses speculative verification during mutual silence, batches already-arrived user frames while the agent listens, and retains dense inference during agent speech; reclaimed GPU time can support resumable retrieval or language-model work. On Moshika Joint, it saves 22.65% of GPU execution time with 99.328% position-wise text-token agreement, not exact conversation outputs.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.