What makes Blabb fast or slow
Where the seconds actually go, why the first dictation after a break is slower, and the levers worth pulling
UPDATED 2026-07-28
From the moment you stop talking to the moment your text is finished typing, expect a second or two for a sentence and longer for a paragraph. Where you land depends mostly on one setting and on whether Blabb has been sitting idle — the first dictation after a break is the slow one.
Where the time goes#
| Stage | What it costs |
|---|---|
| Hearing that you stopped | About 0.7 seconds of continuous silence — Blabb waits for real silence, not a timer |
| Transcription | The bulk of the wait on Original |
| Cleanup | Roughly as much again, and more on the heavier styles |
| Typing | Grows with the length of what you said |
Typing is worth a note because it is the only stage that scales with how much you said. Two of the four text entry methods deliver a character at a time — Type fast and Type slowly — which is invisible on a sentence and quite visible on a 300-word email, text appearing steadily rather than all at once. That is the method working, not a stall. Direct and Paste put the text in in one go: see How your text is entered.
A long dictation does not wait proportionally longer before text starts arriving, because transcription overlaps your speech — segments are transcribed while you are still talking, and the text is held back and delivered when you finish. Only the typing grows with length.
Why the first one after a break is slower#
The cleanup model is unloaded after 15 minutes with nothing to do, and loaded again on your next dictation. So the dictation you do after lunch pays for that load and the ones after it do not. This is deliberate — the alternative is holding a language model in memory all afternoon while you are in a meeting.
The speech model stays loaded, so transcription does not have the same warm-up. It does a short warm-up when it first loads after Blabb starts.
Note
After the PC has been asleep, the first dictation can be slow while models reload. This is a rough edge rather than a design decision. If you have just woken the machine, the second dictation is the one to judge it on.
The levers, strongest first#
Cleanup style is the big one, and it is not close. From Settings → Writing style → Cleanup style:
| Style | Effect on speed |
|---|---|
| Original | No cleanup model runs at all. This is as fast as Blabb gets |
| Fix typos | The lightest style that still runs cleanup |
| Clear, Professional, Friendly, Screen match | Heavier work, more time |
The trade is exactly what you would expect. Original never changes a word, which also means it never fixes one — no punctuation tidying, no grammar, no homophone correction. Fix typos is the usual middle ground.
The specialist vocabularies add a spelling pass over the raw transcription and hand matched terms to the cleanup model. The cost is small, and it buys you correctly spelled drug names or Latin phrases — but it is not free, so leave Transcription mode on General if your work does not need it.
Keeping Blabb running does not avoid the 15-minute unload — that timer runs while the app is open, so a quiet afternoon costs you the reload either way. What it avoids is starting the app from cold on top of it. Settings → General → Start Blabb with Windows makes that automatic.
What does not change the speed#
Speaking more slowly does not help — it makes recognition worse and takes longer. Dictating in shorter pieces does not reduce the per-dictation overhead; it multiplies it. And the length of what you say only affects the typing stage, not the thinking stages.
Next#
- Getting a good result on a modest PC — no discrete graphics, 8 GB of memory, an older laptop.
- Making the most of a fast PC — what a capable machine actually buys you.
- Cleanup styles — what each style does to your words.