Earlybird: a 29M-param speech-to-text model that outputs words before they end
A 17 MB open-weights on-device speech-to-text model that predicts each word correctly a median 90 ms before the speaker finishes saying it — where other streaming models get it right 90–400 ms after — with a confidence score on every word.
Why it matters
It flips streaming STT from reactive to predictive: a voice agent can start acting on a word before it is fully spoken, and at 17 MB on CPU it is small enough to ship inside an app rather than a server.
What you could build with it
An indie developer building a voice agent or live-captioning app could embed Earlybird on-device to cut response latency and keep audio private, starting actions on predicted words instead of waiting for full utterances.
Does it hold up?
Too early to judge — launched within the last day; the arXiv paper and Hugging Face checkpoint are out, but no independent benchmarks or third-party builds yet.
Built with Earlybird
- Earlybird on Hugging FaceHugging Face · The Earlybird-29M checkpoint (sahilmahendrakar/Earlybird-29M) is listed on the Hub under automatic speech recognition, updated within the last day.
- Product Hunt Launches TodayTrendoday · Live Product Hunt feed showing Earlybird at #1 of the day's launches as of 11 Nov 23:42 UTC.
- Earlybird: Predictive Streaming Speech Recognition in 29M Parametersarxiv · The paper: a 29M-parameter streaming recognizer trained on randomly cut-off audio predicts the median word 90 ms before it is spoken, and beats Moonshine Tiny on every Open ASR Leaderboard English set.
- How to Use Earlybird AI for Beginnersaitoolsteps · Beginner's guide: Earlybird launched on Product Hunt Nov 10, is free with open weights and code, runs on an ordinary CPU, and works in English only for now.
Learn more
First spotted on article: source.
More AI models
VoiceStudio: open-source, fully-local ElevenLabs alternative for voice cloning, dubbing, and audiobooksdebpalash/VoiceStudio is a desktop app that runs voice cloning, voice design, video dubbing, dictation, transcription…models · JEV 0.75Edge0-145B: open 145B MoE that runs on a single consumer GPUEdge0 open-sourced 145B and 120B latent-MoE models derived from Qwen3.8-Flash-Next, plus an inference framework that…models · JEV 0.74Strata: run a 125B model on a gaming PCFree, MIT-licensed local inference engine that runs Qwen3.8-Flash-Next (125B MoE, about 6B active per token) on…models · JEV 0.73Google AI Edge Gallery: offline speech recognition and synthesis on-deviceGoogle added fully offline speech recognition and text-to-speech to AI Edge Gallery, with ML Kit APIs on Android and…models · JEV 0.73Soup: fine-tune an 8B LLM on a 4 GB laptop GPU from one YAML fileAn open-source CLI that fine-tunes LLMs from a single YAML file, with layer streaming that trains an 8B model on a 4…models · JEV 0.73Tencent Hunyuan Hy3: open 295B MoE with 21B active under Apache 2.0Tencent open-sourced Hy3, a 295B-parameter mixture-of-experts model with only 21B parameters active per token, under…models · JEV 0.73
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.