# Choose a speech model (/guides/choose-a-speech-model)

This guide changes the speech model that ScribeKey uses on your phone. A speech model turns your recording into text, and all of them work offline once downloaded. Setup downloaded one for your phone's memory. Change it when you want words to appear while you speak, better accuracy, another language, or a smaller download.

[Models](/models) compares every model by language, size and the memory it needs.

## Download and switch models [#download-and-switch-models]

1. Open **Settings**, then **Speech**.
2. Check that **On device** is selected.
3. Tap **Change speech model**. The **Speech models** screen opens on the group of your current model.
4. Choose a group:
   - **Live** shows words while you speak.
   - **Full-size English** writes text after you finish speaking.
   - **Lightweight** has smaller downloads and uses less memory.
   - **Multilingual** covers languages beyond English.
   - **Advanced model library** lists every model, with search.
5. Tap **Model info** on a model to read about it.
6. Tap **Download** on the model you want.

A progress bar shows the download. You can leave ScribeKey. When it finishes, ScribeKey makes it your speech model, and the model shows **Selected**. If it shows **Set active** instead, tap it.

```mermaid
flowchart TD
  accTitle: Activate a downloaded speech model
  accDescr: Download a speech model. If it shows Selected, it is active. If it shows Set active, tap it to make the model active.
  Download([Download model]) --> Status{What does it show?}
  Status -->|Selected| Ready([Model is active])
  Status -->|Set active| Activate[Tap Set active]
  Activate --> Ready
```

The **Speech** screen now shows the model's name.

> **Note:** **Recommended for this phone** marks the model ScribeKey suggests. A model the phone cannot run is listed under **Not supported on this device**, with the reason.

If the download shows **Waiting for Wi-Fi**, tap **Use mobile data**, or turn off [**Download models on Wi-Fi only**](/settings/storage#storage.download_wifi_only).

## Remove a model [#remove-a-model]

1. On **Speech models**, open the model's menu.
2. Tap **Remove download**.

## Add a custom model [#add-a-custom-model]

You can add a speech model that is not in ScribeKey's list, from Hugging Face or from a folder on the phone.

1. On **Speech models**, open **Custom models**.
2. Tap **Add from Hugging Face**.
3. Type a name, or paste a link to a public model, and tap **Find**.
4. Choose a result and tap **Add model**.
5. Tap **Download** on the model in **Custom models**.

To use model files already on the phone, tap **Add from this phone** and choose their folder. ScribeKey never recommends a custom model.

## Related pages [#related-pages]

- [Speech models settings](/settings/speech-models)
- [Dictate in another language](/guides/dictate-in-another-language)
- [On device and cloud](/concepts/on-device-and-cloud)