Reproduce “Earlybird: Predictive Streaming Speech Recognition in 29M Parameters” 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
Sahil Mahendrakar
Earlybird trains a small streaming speech recognizer on audio cutoffs matching live use so it can display complete words from partial sounds with confidence scores. On LibriSpeech test-clean, its median word first appears correctly 90 ms before acoustic end, compared with released streaming models that show words 90–400 ms after it. This measures audio-relative output timing, not end-to-end agent response time.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.