TapWhisper.com

About

TapWhisper is dictation for the whole Mac, not for one app. Hold down fn, say what you want typed, let go: the words appear at your cursor, formatted, in whatever program you were working in. A single tap does the same job hands-free — dictation stays live until you tap again. There is nothing to train and nothing to calibrate. Under the hood, the app types through the accessibility APIs that macOS provides, which is how one utility reaches every text field on the system: no per-app setup, no extension to install, no detour through the clipboard.

The pipeline lives on the laptop. Speech recognition and formatting run on-device, which is why a missing connection changes nothing — dictation keeps working mid-flight and off the grid. Privacy falls out of the same design. The app carries no telemetry, your audio stays put, and so do your transcripts; server-side, the only traces of you are a sign-in and a subscription.

Recognition comes in two flavors. Out of the box, TapWhisper uses the speech engine already inside every Mac — no settings to touch, no download to wait for, and text streams in while you are still talking. Advanced Mode trades that convenience for control: download speech models to run locally, pick the one that suits you, and optionally put a local language model in charge of formatting. A harness inside the app keeps those downloaded models loaded and ready, which makes some of them faster. Whichever engine is active, a local cleanup pass straightens capitalization, turns spoken punctuation into real marks, and strips the filler words. The app detects which of 99+ languages you are speaking without being asked, and it learns your vocabulary — up to 100 terms of your own, plus bundled packs for the jargon-heavy fields: coding, medical, legal, and finance.

The footprint stays small. The download is roughly 14 MB, there is no browser runtime tucked inside, and the target for idle memory is under 50 MB of RAM. TapWhisper is in public beta for Apple Silicon Macs — M1 or newer — running macOS 13 and up. The beta costs nothing to try; paid plans are $5.99 a month or $49.99 a year, both carry a 7-day free trial, and you can cancel anytime.

Highlights

• Push-to-talk typing: fn down to speak, fn up and the words are at your cursor — a single tap toggles it on instead
• Recognition and formatting run on your Mac: nothing you say gets uploaded, zero telemetry, and it keeps working offline
• Two engines: the Mac's built-in recognizer, nothing to configure, words streaming live — or models you download and run locally, with optional language-model formatting
• Detects any of 99+ languages automatically; learns up to 100 of your own terms, with coding, medical, legal, and finance packs bundled
• Roughly 14 MB to download with an idle-RAM target under 50 MB; the public beta is free to try, then $5.99/mo or $49.99/yr with a 7-day free trial

Who it’s for

Mac users who would rather talk than type: developers and professionals dictating into editors, chat, and email; people who refuse to hand audio to a cloud service; offline workers getting through flights and dead zones; and multilingual speakers who move between languages all day. Requires an Apple Silicon Mac (M1 or newer) running macOS 13 or later.

Details