Superwhisper just released S1-mini, a 0.6B LLM designed to turn raw speech-to-text transcripts into clean written text. It can even run 100% locally in your browser on WebGPU, thanks to 🤗 Transformers.js! Try it out yourself! 👇

Congrats to the @superwhisper team on the launch! I can't wait to see what the community builds with it. 🤗 Link to demo (+ source code):

Can't believe it even cuts the gaps where you weren't sure in the demo, amazing power in a tiny model

That's nice. I want it to be transcribing all that i am saying instantly by displaying them while i am saying the words, not when i am done.

0.6B for transcript cleanup is exactly the size that makes sense on device. I build my Local LLM app in SwiftUI with small open-weight models, TestFlight soon. Did you have to fight much with prompt drift at that size, or does the narrow task keep it stable?

Running a 0.6B LLM fully in‑browser is impressive; it lowers latency and privacy concerns for speech‑to‑text cleanup.

what makes s1-mini tick is not just the model size or tech stack but the fact that it tackles one of the last remaining pain points in conversational ai: making raw speech intelligible and editable

@superwhisper I'm not sure what bad I've done but could you kindly unblock me? 🙏🏻

0.6B LLM and mini? for in-browser use?

I tested it! Many mistakes in transcription! Cannot match @NVIDIAAI parakeet-tdt-0.6b-v2

Runs locally in browser, nice

Running a 0.6B LLM fully in the browser showcases impressive WebGPU progress, but expect speed and accuracy limits on typical hardware.

Saw your demo app on hf. Really cool!

something's really off. i'm pretty sure i didn't say "dessert" lmao

0.6b for punctuation is the correct size for once