Transcription
29M-param speech-to-text that outputs words before they end.
Earlybird is a 29M-parameter speech-to-text model that outputs words before they end. It is built for real-time use, where waiting for a full sentence is too slow.
At 29M parameters, Earlybird is small enough to run efficiently and open enough for developers to inspect, adapt and deploy. It trades size for speed, without giving up usable accuracy.
For teams building voice products, transcription tools and real-time AI. Drop Earlybird into a pipeline and get streaming text that starts early and stays aligned.
29M-param speech-to-text that outputs words before they end
See the whole day — 11 November 2026