> ## Documentation Index
> Fetch the complete documentation index at: https://docs.mauvely.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Formatting

> How Sotto turns speech into properly formatted prose: punctuation, paragraph breaks from pauses, spoken commands, and artifact removal.

Raw speech recognition gives you a wall of words. Sotto's job is to make it read like something
you wrote.

## What you get for free

Whisper itself produces **punctuation and capitalisation**. You do not need to say "comma" or
"full stop" — speak naturally and the sentence structure comes out.

## Paragraph breaks from pauses

Pause for longer than a threshold — **2 seconds** by default — and Sotto starts a new paragraph.

This is the feature that makes long dictation usable. Think about your next point, and the break
you took while thinking becomes the paragraph break the reader needs.

Adjust the threshold in **Settings → Formatting**. Shorten it if you speak with lots of small
pauses and want tighter paragraphing; lengthen it if you are getting breaks you did not intend.

## Voice commands

Two spoken commands, on by default:

| Say             | Get               |
| --------------- | ----------------- |
| "new line"      | A line break      |
| "new paragraph" | A paragraph break |

<Warning>
  **Mind the false positives.** "A new line of products" contains the phrase, and Sotto cannot tell
  the difference. If you write about product lines, new paragraphs or newline characters, turn
  voice commands off in **Settings → Formatting** and rely on pauses instead.
</Warning>

## Artifact removal

Transcription models emit markers for things that are not speech. Sotto strips them:

* `[BLANK_AUDIO]` and similar bracketed markers
* Parenthetical noise annotations like "(laughs)"
* Music symbols

## Normalisation

Across the whole dictation, Sotto tidies up:

* Spacing around punctuation
* Capitalisation at the start of each sentence, including where one utterance runs into the next
* Joins between utterances, so a pause mid-sentence does not become a sentence break

That last one matters more than it sounds. Speech gets committed in chunks as you pause, and
without normalisation each chunk would arrive looking like its own sentence.

## Consistency across a long dictation

Each committed utterance is passed forward as context for the next. Names, jargon and technical
terms stay spelled the same way through a long session, instead of drifting between plausible
alternatives.

## What is not done

Sotto formats; it does not rewrite. It will not restructure your argument, fix your grammar, or
change your register — what you said is what you get, punctuated.

An optional local rewriting pass is on the roadmap and would be exactly that: optional, local, and
off by default.

<Tip>
  Sotto works best when you speak as though dictating to a person: complete sentences, a beat
  between thoughts. Talking as if to a search box produces fragments, and no amount of formatting
  fixes a fragment.
</Tip>