which accent the browser recognizer should expect. the local model is english-only and accent-agnostic.
count homophones (to/two/too) and one-letter recognizer slips as correct. strict mode is for people with something to prove.
run speech recognition locally so audio never leaves your machine (newer chrome only).
monkeyap works best with its own speech model: it's more accurate than your browser's, knows dev jargon, and scores you properly at the end of each test.
local mode runs speech recognition on your machine, so nothing you say leaves it. that needs a one-time download of : . after that it's cached and loads instantly.
maybe don't do this on a phone plan.
everyone talks to their coding agent now. but who yaps fastest? monkeyap is a typing test where you don't type. read the words out loud and your browser transcribes them live. you get wpm, accuracy and consistency like any self-respecting speed test.
wpm counts the characters of correctly-said words (plus a space each), divided by 5, per minute. raw counts every word you attempted, right or wrong. acc is correct words over everything you said, skipped or added. the clock starts the moment your voice does, and stops when you stop.
acronyms can be said however you'd naturally say them: "CI/CD" as see eye see dee, "k8s" as kubernetes or kates, "SQL" as sequel, "404" as four oh four.
there's no server. results live in your browser's local storage. speech is transcribed by your browser's built-in speech engine when the engine is set to browser: in chrome that means audio goes to google, safari uses apple's recognizer. set the engine to local and a speech model (moonshine) is downloaded once and runs entirely on your machine, so audio never leaves it.