Formatting
Clean up, rewrite, or translate the transcript before it is inserted.
Formatting is an optional step that runs after transcription. The raw transcript is sent to a language model, and the rewritten text is what gets inserted.
It is configured per profile, so you can have one shortcut that inserts exactly what you said and another that turns the same speech into a polished email.
Presets
| Preset | What it does |
|---|---|
| Normal | Fixes transcription errors — misheard words, homophones, wrong word boundaries — plus grammar, spelling, and punctuation, without changing the meaning. |
| Prompt | Restructures your speech into a clear, well-organised prompt for an AI assistant. |
| Expands the text into a complete email with a greeting, body, and closing, keeping every detail. | |
| Formal | Rewrites in a formal, professional tone with full, polished sentences. |
| Casual | Rewrites in a conversational tone, like a chat message. |
| Romantic | Rewrites in a warm, affectionate tone. |
Every preset is instructed to keep your information and intent intact, add no new facts, and return only the rewritten text.
Translate
Translate rewrites the transcript into another language. Just under a hundred languages are available, so you can speak in one language and have the text arrive in another.
This is separate from the profile's language setting, which describes the language you speak.
Custom formatting
Custom lets you write the instruction yourself, in one of two ways:
- Task (default) — describe what you want, for example
translate to Frenchorturn this into bullet points. Your text is wrapped in OiPer's formatting template, which supplies the surrounding rules about preserving meaning and returning plain text only. - Advanced mode — your text is used directly as the full system prompt, with no template around it. Use this when you need complete control over the model's behaviour.
Which model runs it
In local and cloud modes, formatting runs through an LLM you registered on the
Models page; if none is set, formatting is skipped. In oiper mode
it runs on OiPer's hosted models with nothing to configure.
Formatting adds the model's response time to every dictation. A small, fast model keeps the delay short; a larger one gives better rewriting.