# Speech (/settings/speech)

Speech decides how your recording becomes text. Open it to choose between a speech model on your phone and a cloud service, to set the language you speak, or to change when a recording stops.

At the top of the screen, **On device** turns your speech into text with a speech model on your phone, and the audio never leaves it. **Cloud** uploads your audio to a cloud service you set up. **Change speech model** opens [Speech models](/settings/speech-models).

Open it from **Settings › Dictation › Speech**.

## Spoken language [#speech.spoken_language]

The language you speak. With **On device**, the choices depend on the speech model: a model that takes one language shows that language here and offers no choice. With **Cloud**, you can pick a language or enter another language code. **Auto detect** lets the model or the service work out the language, which suits you when the language you speak may change.

## Cloud provider [#speech.cloud_provider]

Shows the cloud service and model that **Cloud** uses, or what is still missing, such as an endpoint or a key. Tap it to set up the connection. This row appears while **Cloud** is chosen. Cloud speech sends your recorded audio, the model name, and the spoken language to the service you set up. See the [privacy policy](https://scribekey.app/privacy).

> **Note:** While [**On device only**](/settings/privacy#privacy.on_device_only) is on, ScribeKey pauses this control and says why. Turn that setting off to use it.

## Pause music [#speech.pause_music]

Stops music and other audio while you dictate. It is on by default.

## Stop after silence [#speech.stop_after_silence]

When this is on, a recording ends by itself when you stop speaking. Short pauses inside a sentence are kept. A slider then sets how long a silence ends the recording, from 0.5 seconds (**Responsive**) to 3 seconds (**Patient**). The default is 1.2 seconds. It is off by default, so a recording runs until you stop it.

## Keep speech ready [#speech.keep_ready]

Keeps the on-device speech model loaded between dictations, so a dictation from the bubble starts faster. It applies only while **Show the bubble** is on. It is on by default.