---
title: How dictation works
description: Hold the hotkey to listen, speak, and paste a transcript into the focused field — or copy it if paste is blocked.
url: https://pr-1-2b0a782aae90.thally.app/dictation
---

# How dictation works

Hold the hotkey to listen, speak, and paste a transcript into the focused field — or copy it if paste is blocked.

Polyflo is push-to-talk. Nothing is recorded until you hold the hotkey, and processing starts only when you release it.

## The loop

#### Focus a text field

    Click into any application that accepts typed text. Polyflo remembers that window so it can return focus after transcription.

#### Hold the hotkey

    Default shortcuts:

    - Windows: `Ctrl+Shift+Space`
    - macOS: `Cmd+Shift+Space`

    Change them in [Settings](/settings). Holds shorter than about 180 ms are ignored so accidental taps do not start a session.

#### Speak

    A circular overlay appears and shows a live waveform while you talk. Keep holding the hotkey until you finish the phrase.

#### Release

    The overlay switches to processing. Polyflo sends the recording to Sarvam, then pastes the result into the original field.

> **Note:**
  If the Sarvam key is missing, dictation does not start. Settings opens (or a notification appears) so you can add one. See [API key](/api-key).

## Modes

Choose a mode at the top of Settings. It applies to the next dictation.

| Mode | Label in Settings | Result |
| --- | --- | --- |
| **Transcribe** | Transcribe | Speech is recognized with automatic language detection and pasted as spoken. |
| **Translate** | Translate | Speech is transcribed and translated to English, then pasted. |

Transcribe uses Saaras v3 `transcribe` mode. Translate tries Saaras `translate` mode first, then falls back to transcription plus Sarvam Translate if needed.

## What happens to the text

1. The transcript is written to the system clipboard.
2. If this platform can simulate paste, Polyflo restores the original window and sends paste (`Ctrl+V` / `Cmd+V`).
3. If paste is unavailable or fails, the text stays on the clipboard and a notification tells you to paste it yourself.
4. The transcript is stored in Settings → **Clipboard** (up to 50 recent entries).

## Overlay states

| State | What you see |
| --- | --- |
| Idle | No overlay (or a tiny idle marker) |
| Listening | Expanded circle with a live audio waveform |
| Processing | Spinner while Sarvam returns a transcript |

The overlay stays above other windows and does not steal keyboard focus.