Reproduce “Latency Control for Keyword Spotting” 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
Christin Jose, Joseph Wang, Grant P. Strimel, Mohammad Omar Khursheed, Yuriy Mishchenko, Brian Kulis
The authors introduce a tunable training method for conversational keyword spotting that shifts positive-example supervision from the highest-scoring frame to the preceding frame with probability b. This encourages earlier detections without requiring accurate endpoints as training targets. In binary CNN detection of “Amazon,” they report a 25% relative false-accept improvement at fixed latency over a max-latency baseline, in this setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.