0 words
Whisper Dictation types out English speech as you talk, with the Whisper model (Tiny, Base, Small or Large) running in your browser on WebGPU or WebAssembly. The audio and the words never leave the browser, and earlier sentences are transcribed while you are still speaking, so the text is nearly done when you stop.
The chosen model starts loading when the page opens and the browser keeps it for next time.
Tiny is the fastest and the first choice on a first visit. Large (Whisper large-v3-turbo) is the most accurate: a 1.5 GB download that needs a graphics card, so it is not offered on phones.
0 words