> For the full page index, fetch https://echonova.ai/llms.txt

# Dialects

The dialects the voice library carries, how the library groups voices by them, what a dialect changes in generated speech and in an agent, and how to pick one for a region.

A dialect is the accent a voice speaks in, and every voice in the library
speaks exactly one. Read this page to decide which dialect to give a
narration or an agent for the people who will hear it, and to know what that
choice changes and what it leaves alone.

## A voice speaks one dialect

You never choose a dialect on its own. You choose a voice, and its dialect
comes with it. That is why the library shows the dialect on every row, and why
an agent that must speak two dialects needs two voices.

The library can carry voices in these dialects:

| Dialect      | What it is             | Language |
| ------------ | ---------------------- | -------- |
| **Egyptian** | Egyptian Arabic        | Arabic   |
| **Emirati**  | Emirati Arabic         | Arabic   |
| **Saudi**    | Saudi Arabic           | Arabic   |
| **Kuwaiti**  | Kuwaiti Arabic         | Arabic   |
| **Qatari**   | Qatari Arabic          | Arabic   |
| **Omani**    | Omani Arabic           | Arabic   |
| **Lebanese** | Lebanese Arabic        | Arabic   |
| **Syrian**   | Syrian Arabic          | Arabic   |
| **Iraqi**    | Iraqi Arabic           | Arabic   |
| **Fusha**    | Modern Standard Arabic | Arabic   |
| **American** | American English (AM)  | English  |
| **British**  | British English (BR)   | English  |

**Fusha** is Modern Standard Arabic: the Arabic of news bulletins, formal
announcements and written prose, understood across the region. Every other
Arabic dialect is the speech of one country.

Older English voices may still show **English** as their dialect.

Which dialects have voices at a given time can change. The library is the
source of truth: see the next section.

## The library groups voices by dialect

The dialect is the main way into the library, so the same grouping appears in
three places.

- **The chip row** under the search box on **Library**. Each chip shows how
  many voices the dialect has. Dialects with no voices yet follow with the
  word **soon** and cannot be pressed. Press a chip to see only that dialect.
- **The voice rows**, which are grouped by dialect. A dialect with no voices
  yet shows a line such as **Egyptian · coming soon**.
- **The dialect chips on Text to Speech** and in the voice picker. Press a chip
  to see that dialect's voices.

The dialect a voice shows in the library is the dialect it speaks everywhere
you use it: on **Text to Speech**, on an agent and through the Developer API.

## What a dialect changes when you generate speech

The dialect fixes the accent of the audio, and nothing else about the request.

- Picking a voice picks the dialect. When you arrive on **Text to Speech**
  through **Use in Text to Speech**, the page sets the dialect to match the
  voice you chose. Emirati is the dialect the page starts on.
- The voice reads your text in its own accent whatever the text's language.
  Arabic text on an English voice comes out with English pronunciation. Pick
  the voice for the language the text is written in.
- The **Tashkeel** switch appears under the writing area while an Arabic
  voice is selected, whatever its dialect. With an English voice it is not
  shown. See [Add diacritics](/docs/playground/add-diacritics).
- The options are the same for every dialect. **Speed**, **Guidance** and
  **Quality** have the same ranges and defaults whichever voice you pick.

## What a dialect changes for an agent

An agent carries one or more voices, and each brings its dialect with it.

- The **default voice** is the accent the agent opens with. Set it on the
  agent's **Voice** tab.
- [Dialect detection](/docs/agents/system-tool-dialect-detection) lets the
  agent recognise the caller's Arabic dialect and switch to the matching voice
  among the voices it carries. An agent with one voice keeps that accent for
  the whole conversation.
- **Tashkeel** is offered in a voice's **Voice Settings** on every Arabic
  voice. It is off by default.
- A text conversation is not spoken, so the dialect of the agent's voices has
  no effect on WhatsApp, Slack, a helpdesk or Genesys chat. The language the
  agent writes in comes from its persona, not from a voice.

## Get it right

Match the dialect to the listener, not to the text you happen to have or to
where your company is based. A caller hears the accent before the words, and
an accent from the wrong place costs you the first seconds of every call.

**Recommended**

- One region: give the agent a voice in that region's dialect and make it the
  default. Cairo callers get an **Egyptian** voice; Jeddah callers get a
  **Saudi** voice.
- Several regions: add one voice per region to the agent and turn on dialect
  detection, so each caller is answered in their own accent.
- A formal read for the whole region, such as an announcement, a product film
  or a news-style narration: a **Fusha** voice, with **Tashkeel** on
  the text.
- English text: an **English** voice.

**Less effective**

- One **Fusha** voice for a casual support line in Cairo. Callers understand
  it, and it sounds like a broadcast rather than a conversation.
- One Gulf voice for callers across the whole region.
- Arabic text read by an English voice, or English read by an Arabic voice,
  because that voice sounded best in the preview.
- A line in the persona asking the agent to "sound Egyptian" on a Saudi voice.
  The words change; the accent does not.

## Worked example

Acme sells in Cairo and Jeddah. Its support agent gets two voices from the
library: an **Egyptian** voice made the default and a **Saudi** voice, with
dialect detection on. A Cairo caller is answered in Egyptian from the first
word; a Jeddah caller hears the Saudi voice from the turn the dialect is
recognised.

For the product film that runs in every market, the same team opens **Text to
Speech**, picks a **Fusha** voice, turns on **Tashkeel** for the script and
generates one narration that works in Cairo and Jeddah alike.

## FAQ

### Is Fusha a dialect?

Fusha is Modern Standard Arabic rather than the speech of one country, but the
library lists it beside the dialects because you choose it the same way: by
picking a Fusha voice.

### Can one voice speak two dialects?

No. A voice has one dialect. To cover two, add a voice for each.

### Which dialects have voices right now?

Open **Voices**. The chips you can press are the dialects that have voices
today, and each chip shows its count. The ones marked **soon** have none yet.

### Do I choose a dialect or a voice on Text to Speech?

Choose the voice. Its dialect comes with it, and the page shows that dialect
as the one in use.

### Does Tashkeel work on an Egyptian voice?

Yes. It is offered on every Arabic voice, not only on **Fusha**.

### Does the dialect change what the agent says?

No. The dialect sets the accent. What the agent says, and in which language,
comes from its persona and prompt. See
[Voices and language](/docs/agents/voice-and-language) for how to write the
language rule.

## Related

- [Browse the voice library](/docs/playground/voice-library): the chips, the voice rows and the details panel
- [Filters, previews and favourites](/docs/playground/filters-previews-favourites): filter by dialect and compare voices
- [Choosing a voice](/docs/playground/choosing-a-voice): what to listen for once the dialect is decided
- [Voices and language](/docs/agents/voice-and-language): add a voice per dialect to an agent and set the default
- [Dialect detection](/docs/agents/system-tool-dialect-detection): switch accent mid-conversation
- [Text to Speech voices and options](/docs/developers/tts-voices-and-options): the dialects as the API sees them