# Live captions

The caption window shows what Blabb is hearing while you speak, so you know you are being picked up.

While you dictate, Blabb shows what it is hearing, as it hears it. That is the point of the
caption window: to answer one question, in the moment, without you having to stop and check
anything.

## The question it answers

Am I being heard?

If your words are appearing, the microphone is working, Windows is letting Blabb use it, and
the speech model is picking up your voice. You can carry on. If the captions stay empty while
you talk, something upstream of the text is wrong — the microphone is muted, the wrong device
is selected, or you are too far from it.

That is genuinely all it reports. It is not a progress bar, not a status panel, and not a
place errors are explained. When a dictation fails, the seven dots on the pill turn red and
the log records why — see [The overlay pill](/docs/get-started/the-pill/).

## The one exception

If Windows is blocking microphone access, the captions say so, because that is the one problem
that stops you dead and cannot be worked out from an empty window. Blabb also opens the Windows
microphone settings so you can turn access back on. The fix is in
[Microphone blocked](/docs/common-issues/microphone-blocked/).

## Why the caption does not match the final text

This is expected, and it is worth understanding rather than reporting as a bug.

The captions show hearing. They are the raw transcription — what the speech model made of your
voice, including your "um"s and your restarts.

The final text has been through whichever cleanup style you chose. So the caption is the earlier,
rougher version by design.

What changes between the two depends entirely on that style. On **Original** the answer is
"nothing" — the captions and the typed text match. On the others:

| In the captions | In the typed text |
| --- | --- |
| "um", false starts, repeated words | Removed by Clear, Professional and Friendly. Fix typos deliberately keeps them |
| "new paragraph" as spoken words | An actual blank line and a new paragraph |
| Rough grammar and casing | Tidied |
| Plain wording | Rewritten if you chose a style that changes tone |

How far apart the two versions look depends entirely on your cleanup style. **Original** runs
no cleanup model at all, so the typed text is closest to what you saw in the captions.
**Professional** and **Friendly** rewrite the wording, so they are furthest away. The full
picture is in [Cleanup styles](/docs/cleanup-styles/overview/).

> [!TIP]
> If the captions look right but the typed text does not, the problem is cleanup, not hearing —
> and the cure is a lighter cleanup style, not a better microphone. If the captions look wrong,
> it is the other way round.

## When the captions are wrong

Wrong words in the captions mean the speech model misheard you, and no setting downstream will
fix that. The things that actually help are microphone placement, a quieter room, and speaking
at a natural pace rather than over-enunciating. Start with
[Microphone setup](/docs/accuracy/microphone-setup/).

For names, products and terms Blabb reliably gets wrong, teach it the spelling once with
[Saved words](/docs/cleanup-styles/saved-words/).

## Next

- [Cleanup styles](/docs/cleanup-styles/overview/) — what happens between the caption and the text.
- [When words come out wrong](/docs/accuracy/when-words-come-out-wrong/) — fixing recognition itself.
- [Your dictation history](/docs/get-started/history/) — compare the raw words and the final text after the fact.