# Speech

> Turn written text into speech on a canvas or in chat, pick one of Endless' voices or your workspace's, and make a new voice by describing or cloning it.

Source: https://endless.app/docs/create/speech

**Speech** reads text you write out loud. You pick a **voice** (a name and a description), and
Endless knows which model speaks it. There is no separate page: you make speech on a
[canvas](/docs/create/canvas) or in [chat](/docs/chat), from the same list of voices.

## Make speech on a canvas

<Steps>
  <Step>
    ### Enter the text

    With nothing selected, choose &#x2A;*"Generate speech"*&#x2A; and type in &#x2A;"Write the text you want to
    turn into speech…"&#x2A;. What you type is exactly what is read. To use text already on the board,
    select the text shape and choose &#x2A;*"Convert to speech"**.
  </Step>

  <Step>
    ### Pick a voice

    The model list becomes a voice list. If a voice can be spoken by more than one model,
    &#x2A;*"Model"*&#x2A; under &#x2A;*"Advanced parameters"*&#x2A; switches it; &#x2A;*"Reset all"** goes back to the
    voice's own.
  </Step>

  <Step>
    ### Check the estimate and run

    The estimate prices your text. The result is an audio clip on the board, also saved to your
    library.
  </Step>
</Steps>

**"Generate speech"** appears only when your workspace has a speech model turned on.

## Voices

* **Endless voices** are ready in every workspace.
* **Workspace voices** are made by members. A new voice is private to its maker until they
  share it with the workspace.

## Make a new voice

**"Create new voice"** is at the end of the voice list (and shown when a search finds nothing).
It opens the same form as &#x2A;*Settings → Voices → "Create voice"**:

| Option                           | How it works                                                                                                                                                    |
| -------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Create by prompting**          | Describe age, accent, tone and pace. Ready in under a minute and uses no voice slot.                                                                            |
| **Create from audio references** | Clone one person from recordings with one speaker and no music or noise. You must confirm the person agreed. A workspace has a limited number of cloned voices. |

<Callout type="warn" icon="<Coins />">
  **A described voice costs a few extra credits every time it speaks**, because
  its sample is generated again. The estimate includes it.
</Callout>

Making a voice needs permission to create content. In **Settings → Voices** you share a voice,
make it private again or delete it. When the clone slots are full, delete one there or describe
a voice instead.

## Speech in chat

Ask chat to read something out loud and name the voice. Chat can also design a voice from a
description or clone one from recordings you attach. A clone always stops on a confirmation
card so that you, not the agent, confirm consent. Scheduled runs can't clone.