By Denys Medvediev

Troubleshooting

Dictation cuts off your last words? Here's the fix

Dictation that clips the end of your sentence is almost never a broken microphone — it is an auto-stop firing on your trail-off. The fix is to disable that stop, hold a push-to-talk key, or use a tool with no silence timeout.

Last updated: June 2026

Sound wave on a dark audio interface, illustrating dictation losing its final words

Dictation cuts off the end of sentences when the tool treats a short pause as "you're done" and closes the recording before your final words are spoken. Built-in dictation on Mac and Windows auto-stops on silence, so trailing-off words get clipped. The fix is to disable that auto-stop, hold a push-to-talk key, or use a tool with no silence timeout.

You say the whole sentence. The screen shows all of it except the last three words. I've watched this happen mid-thought more times than I can count — you trail off softly, the way people actually talk, and the software decides the quiet bit was the end.

The boring truth is that almost every built-in dictation tool guesses when you've finished by listening for silence. Your trail-off looks exactly like silence. So it stops, and the tail of your sentence never makes it in. Keep this narrow: if dictation quits a few seconds in and nothing more transcribes, that's a different fault. This article is only about the last few words getting clipped while the rest transcribes fine.

Why your last few words vanish

Studio microphone against a dark background, evoking a noise gate clamping quiet audio

There are four reasons the tail of your sentence disappears. Most people hit the first one and never realize the other three exist. One: the tool auto-stops when it hears silence, and your pause to think tripped it. Two: you released a push-to-talk key a beat too early, before you'd finished the word. Three: your microphone's noise gate cut the quiet, low-energy syllables at the end — the soft consonants that fade out. Four: on cloud dictation, the last chunk of audio was still crossing the network when the stream closed, so it never came back as text.

The first cause is the big one. The other three are the gaps nobody writes about. The fix is different for each.

The silence timeout is firing too soon

Old-fashioned alarm clock on a desk, representing an over-eager stop timer

Most built-in dictation listens for a stretch of silence and treats it as the end of your sentence. Pause to find the right word, and the recording closes. Whatever you say after that pause never gets transcribed.

On Mac, this is the default. Apple's Dictation auto-stops when it detects no speech for about 30 seconds. People read that as a "30-second limit," but it isn't a cap on how long you can talk — Apple's own words are that you can dictate text of any length without a timeout. It's the silence that ends it. Keep talking and it keeps going. Go quiet to think, and the detector decides you're finished.

You can change this. In System Settings, open Keyboard, then Dictation, and turn off the option that auto-ends when you stop speaking. With that off, dictation keeps listening through your pauses instead of bailing the moment you breathe. Apple documents the toggle on their Use Dictation on Mac page. To verify the fix: dictate a sentence, pause a full three seconds in the middle, then finish it. If the whole sentence lands, the timeout was your culprit.

Windows clips your tail on a pause too

Windows does the same thing, just faster. Win+H voice typing automatically pauses when you stop for a moment to think. A Microsoft Community Support Specialist confirmed this is by design — it also pauses the instant you type manually on the keyboard. Users report the pause kicks in after a few seconds of silence, which is plenty to clip the end of a slow sentence. Win+H also needs an internet connection to work at all.

There's a "Wait time before acting" setting in the Win+H toolbar, and the internet will tell you it gives slow speakers more room. It doesn't. Microsoft's own page is clear that this setting controls how long Voice typing waits before carrying out a voice command — not how long it listens before pausing on silence. Tuning it won't stop your trailing words from dropping. The honest fix on Windows is the same as everywhere: don't pause for long inside a sentence, or use a tool that doesn't pause on silence at all. Microsoft's voice typing guide covers the basics. If Win+H drops out completely rather than clipping your tail, that's the voice-to-text not working on Windows problem instead.

Microsoft Word eats your tail after a few seconds

Word's built-in Dictate has its own version of this. It's widely reported to switch itself off after a few seconds of silence, dropping whatever you were about to say next. I say "reported" on purpose — Microsoft's official Word dictation page doesn't document any silence timeout or state a number, so the often-quoted figure is a community thread, not an official one.

The forum workarounds are what you'd expect from frustrated people. Some make a quiet filler sound to keep the session alive while they think. Others dictate in Win+H or Google Docs and paste the result into Word. Both work. Both are also a bit absurd — making an "mmm" noise at your laptop so it doesn't think you've fainted is not what 2026 was supposed to feel like. If you're humming to your word processor, that's the software's bug, not yours.

The cause nobody names: you let go of the key too early

Close-up of hands on a black keyboard, evoking when to release the hotkey

If you use a push-to-talk tool — hold a key, speak, release — there's a cause the OS-dictation articles never mention. You can clip your own tail by releasing the key half a beat too soon. The recording stops the instant you let go. Let go on the last word instead of just after it, and that word gets cut.

This is the honest part of the topic. Releasing push-to-talk a beat too early, or pausing long enough to trip an auto-stop, drops your trailing words in any tool. No software can transcribe audio it never recorded. The fix is technique: hold the key until you've finished, then hold it a moment longer — count a soft "and" in your head — before you release. It feels silly the first three times. Then it's muscle memory and you stop losing endings. I built a push-to-talk tool and still catch myself letting go early, so I'm not preaching from a clean conscience here.

Noise gates and quiet endings

The end of a sentence is its quietest part. You start strong and fade out — that's how speech works. A microphone noise gate, or an aggressive voice-activity filter, can read those soft final syllables as background hiss and mute them. The words were spoken; the mic decided they weren't loud enough to count.

Two fixes. Raise your microphone input level so the quiet tail clears the threshold, and record somewhere quieter so the gate isn't so aggressive to begin with. This isn't our invention — Google's troubleshooting tells users seeing "We're having trouble hearing you" to move to a quiet room, plug in an external microphone, and adjust the input volume. Their voice typing help page has the steps. To test it: say a sentence that ends on a soft word like "yesterday," at your normal fade-out volume. If the last word now lands, your mic was the gate.

Cloud dictation on slow Wi-Fi loses the last chunk

Abstract orange network of lines and dots on dark, evoking cloud transcription lag

Some dictation runs in the cloud. Your speech is streamed to a server, transcribed, and sent back — Win+H needs internet to work at all. When the connection is slow or drops for a second, the final chunk of audio can still be mid-flight when the stream closes, so its text never arrives. The middle of your sentence transcribed fine because the network was healthy then. The end vanished because the last packet didn't make the round trip.

You can't fix network latency from a settings panel. What you can do is run transcription locally, with no server in the loop. Whisper's local mode runs entirely on your own computer and works completely offline — no internet during transcription — so there's no round trip and no last chunk to lose. The only time it touches the network is the one-time model download. After that, your audio never leaves the laptop, and neither do your last three words.

How hold-to-talk keeps your tail

Cancel
Whisper's recording overlay during a held-key session — it records until you let go, with a brief tail buffer so your final words aren't clipped.

Built-in dictation clips your endings because it's guessing when you're done from silence. Take away the guessing and the clipping goes with it.

Whisper records while you hold the hotkey and transcribes the whole captured audio when you release. There's no silence-based auto-stop — a pause never decides you're finished, because you do, with the key. The default is Ctrl+Space on Windows and the Command+Option chord on macOS; hold it, talk, release. And because people naturally trail off, releasing the key the instant you stop would still risk clipping the soft last syllable. So after you let go, Whisper keeps the microphone open for a brief tail buffer, just long enough to catch the words that fade out.

I'll be honest about the limit, because the voice files I write to demand it. The tail buffer forgives a natural trail-off. It does not forgive letting go of the key a full second early — release too soon and you'll still drop the tail, same as any tool. Hold until you've finished, give it that extra beat, and the endings stay put.

When it's not the dictation — fix the right thing first

Before you blame the dictation engine, rule out two things. If your last words drop only inside one specific app — fine everywhere else, broken in one — that's likely the app eating focus or input, not the transcriber. Whisper pastes system-wide wherever your cursor sits, not into one tab or window, so that test isolates it quickly. And if the tail only drops on slow Wi-Fi with a cloud tool, the network is the problem. Fix the connection, or go local, before you change anything else.

And if you're only dictating short bursts, you may not need a third-party tool at all. Apple Dictation is free, built in, and fine for a 30-word text. Turn off its auto-end-on-silence so it stops clipping you, and for short notes it does the job. We start being worth the switch when you're dictating longer than a paragraph, pausing to think, and tired of losing the ends of sentences. Below that threshold, the built-in tool with the right toggle flipped is the honest answer.

The last word is usually the one you cared about most — the verb that finishes the thought, the name you were getting to. Losing it is maddening precisely because the rest landed fine. Flip the toggle, hold the key a beat longer, or hand the job to something that doesn't guess when you're done. My younger daughter dictates emails to her grandmother and has never once lost the ending, mostly because she has no idea you're allowed to pause.

Whisper
The real Whisper app — Local plus Cloud, push-to-talk, no silence timeout. Click around the Settings; it's live.

Want dictation that keeps your last word?

Download Whisper, hold the key until you mean it, and watch the whole sentence — tail and all — land at your cursor.

Photo of Denys Medvediev

Denys Medvediev

I'm the one who reads our support email, most probably by dictating the replies.