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

# Dictation

> Voice-to-text that works in every app. Resonant delivers transcribed text directly into the active text field with smart punctuation and formatting.

## How it works

When you press the trigger key, Resonant begins capturing audio from your microphone. Speech is transcribed locally using a neural speech model running on your device's hardware accelerator (Apple Neural Engine on Mac, CPU/GPU on Windows).

The transcribed text is delivered directly into whatever text field is currently focused — no copy-paste required.

## Voice models

Resonant ships with and supports multiple speech-to-text models. You can switch models in **Settings → Models**.

Models run entirely on-device. Larger models are more accurate but use more memory and may have slightly higher latency on older hardware.

## Smart formatting

Resonant automatically handles:

* **Punctuation** — periods, commas, question marks, and exclamation points
* **Capitalization** — sentence-start and proper noun capitalization
* **Numbers** — "twenty-three" becomes "23" via inverse text normalization (ITN)
* **New lines** — say "new line" or "new paragraph" for line breaks

## Trigger key

The default trigger key varies by platform:

| Platform | Default                       |
| -------- | ----------------------------- |
| macOS    | `Fn` (Globe key) double-press |
| Windows  | `Ctrl + Shift + Space`        |

You can change the trigger key in **Settings → Input**.

## Cloud Cleanup (Pro)

Pro subscribers can optionally send transcripts to an LLM for grammar correction, filler-word removal, and sentence restructuring. This is an opt-in feature — the raw audio never leaves your device, only the text transcript is sent for cleanup.

## Acoustic biasing

Resonant reads text visible on your screen (via OCR) to bias the speech model toward domain-specific vocabulary. This improves accuracy for technical terms, project names, and jargon that the base model might not recognize.

<Note>
  Acoustic biasing requires the Accessibility permission on macOS.
</Note>