Why dictation eats your first word

You press the key and start talking — “Morning team, quick update on the launch” — and the transcript comes back missing its opening: “team, quick update on the launch.” Sometimes it’s subtler than that. You get “orning,” half a word, clipped like a bad phone line.

If you came here searching for why that happens, here’s the honest answer: it’s probably not your diction, and it’s probably not the speech model. When you pressed the key, the microphone was asleep. Waking it takes a beat.

Your first word happened during the beat.

The gap between pressing and listening

When no app is recording, your Mac doesn’t keep the microphone running. Why would it? So when a dictation app says “start listening,” there’s real work to do first: the system claims the device, powers it up, and agrees with it on a format. All of that is fast — a fraction of a second — and for most software, a fraction of a second is free.

Speech is the exception. Talking starts instantly. By the time your finger finishes the press, the first syllable is already out of your mouth. The hardware is still stretching. The syllable lands in the gap, and the gap doesn’t keep anything.

Here’s the part you might not have noticed: you already learned the workaround. Watch anyone who dictates daily and you’ll see it — press… pause… speak. A polite little beat before the first word. Nobody taught them that. The chopped words did.

That’s worth sitting with for a second. The tool trained you to wait for it. You’ve been holding the door open for the hardware, dozens of times a day, and calling it a habit.

A mic that’s on, with nobody home

The obvious fix is to not let the microphone fall asleep — keep the device open while you’re idle, so pressing the key skips the wake-up entirely.

And the moment I say that, you should feel a hair go up on your neck. An always-open microphone is exactly the thing a privacy-first app is supposed to be against. Is it listening?

So here’s precisely what Clabrate’s version does, because anyone proposing a warm mic owes you precision. The setting is called Warmed-up mic, and it lives in Settings → Audio. While you’re idle, it holds the audio device open — and the sound goes nowhere. Each slice of audio is discarded the moment it arrives. It never reaches transcription, never touches your history, never gets written anywhere. There’s nothing behind the microphone but a drain.

The mic is on, but nobody’s home.

Then you press the dictation key, and what flips isn’t power — it’s attention. The idle stream is dropped, real capture starts on a device that’s already awake, and your first word arrives with all of its letters. The switch you always wanted was never “off to on.” It was “not listening to listening.”

The honest cost

There are two, and you should know both before you flip the toggle.

First: macOS shows its recording indicator whenever any app is holding the microphone. With Warmed-up mic on, that little orange dot stays lit while Clabrate sits idle — because the device really is open. We could pretend that’s awkward. Actually, we’re glad it works that way. The operating system refuses to keep secrets about microphones, and a company that found the dot inconvenient would be telling on itself.

Second: holding a device awake costs some battery. Not much, but not nothing, and pretending otherwise would be the kind of rounding-up this blog tries not to do.

That’s why the setting ships off. A warm microphone should be a decision you make on purpose, understanding the trade — never a default you discover later by noticing a dot in your menu bar. If the clipped first word costs you more than the battery does, the toggle is in Settings → Audio, and it takes effect the moment you flip it.

Before you change anything, though, try the experiment that convinced me. Tomorrow, watch yourself dictate. Notice whether you make the pause — press… beat… speak. If you don’t, leave the setting alone; you’re not paying the tax, and off is the right default. But if you catch yourself waiting for the machine, flip it on for a day and notice what disappears. Not just the clipped words. The wait itself. You stop granting the hardware its little courtesy, because there’s nothing to be courteous to anymore.

That’s the part I’ve come to love about this feature, and it’s bigger than the feature. A tool that listens should never train you to wait for it — the pause belongs to your thinking, not to a device driver. The first word of a thought is usually the one carrying the momentum, and it’s yours.

“Morning team” should arrive with its M.