Megaphone

Megaphone streams your words as you speak rather than waiting for you to stop. It leans on Apple's own SpeechAnalyzer, so the heavy lifting happens on the device instead of a remote server. Speed and privacy, same package. Nothing you say leaves your Mac.

Raw dictation is messy. Megaphone's Smart mode runs your transcript through an on-device language model that removes filler words and false starts, turning "Um, I I think we should ship Thursday, no, Wednesday" into a clean sentence. There's a deterministic fallback so cleanup still works if the model can't keep up, plus Basic and Exact modes when you want less intervention. Not ideal if you like full control, but useful if you don't.

You can teach Megaphone the names, products, and acronyms that show up in your writing. Once a term is saved, both transcription and cleanup spell it your way. Automatic Learning picks up words you use often and promotes them after three successful dictations, and you can import or export the whole Dictionary as a JSON file to move it between Macs. Trust me, that export button saves a lot of retyping.

Megaphone reads the active app and window as local hints. It uses that context to judge tone, spelling, and technical formatting, so dictating into a code editor treats --force-with-lease differently from a line in an email, keeping flags intact instead of guessing that a spoken phrase was a typo. Nothing about that context leaves the machine.

Starting a dictation with "Hey Megaphone" hands your words to the on-device Foundation Model, which answers inside the app you're already using, so you can ask it to shorten a paragraph or rewrite a release note without ever opening a chat window or bouncing text through the clipboard.

Megaphone keeps a running memory of your style and recurring terms locally, so it gets more accurate the longer you use it. That data stays on your Mac. It travels with you only if you export it yourself.