Reproduce “Decode Before You Hear: Speculative Listening for Chunk-Interleaved Full-Duplex Speech Language 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
Yutong Bian, Yanglin Zhang, Yuxiang Wang, Luoyuan Zhang, Rongshen He, Zihao Wang, Junwen Qiu, Jiaqi Li and 1 more
Speculative Listening uses a chunk-interleaved voice model to draft during idle listening, assuming silence, then verifies the draft when audio arrives. Across four models, it yields 2.6–2.8× average speak-chunk speedups and up to 23.41× on Streaming SLM tool-call chunks. These are Thinker post-audio latency results, not measurements of total computation or speech synthesis.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.