Arabic text to speech in n8n: turn any workflow into a Gulf Arabic voice

n8n is where a lot of Gulf businesses glue their tools together: a new order, a form submission, a daily report. Adding a voice to those flows used to mean English-only services or a broadcast-MSA voice nobody in Dubai talks like. This guide shows how to add Arabic text to speech in n8n with the built-in HTTP Request node — no community node, no code.

What will the workflow do?

A typical flow:

1. Trigger — a webhook, a schedule, a new row in a sheet.
2. Text — build the Arabic sentence, from a template or a language model.
3. HTTP Request — send it to nutq, get a WAV back.
4. Deliver — post the audio to Telegram, send it by email, or save it to storage.

Step 3 is the only new part.

How do you set up the credential?

In n8n, create a credential of type Header Auth:

Store it as a credential and select it in the node. Never paste the key inline: exported workflows and screenshots leak inline values.

How do you configure the HTTP Request node?

The JSON body:

{
  "text": "{{ $json.message }}",
  "voice": "rashid"
}

voice is any Arabic voice from GET /v1/voices, or one you saved on your account. The node's output is a binary file — a WAV — that the next node can attach or upload. Usage comes back in the response headers (X-Nutq-Characters, X-Nutq-Cost-USD), so you can log what each run cost.

Can the workflow use several voices and pauses?

Yes, in the same request. Open a line with a voice name in square brackets and that voice speaks until the next one; [pause] adds a gap:

[Rashid] شحالك؟
[Mariam] بخير الحمد لله
أهلا بكم [pause:800] ونبدأ

A bare [pause] is 500 ms, [pause:2s] is two seconds, and silence is not billed. Each line is generated on its own and joined, so it sounds like two people reading lines rather than an unscripted conversation — right for announcements and short dialogues.

How do you get the pronunciation right?

Arabic is written without short vowels, so an ambiguous word is a guess. If a word comes out wrong, add diacritics to that word only — marking every word makes the delivery slow and formal. The details, with examples, are in Arabic text to speech with voice cloning.

How do you handle errors in the workflow?

Turn on the node's Continue on fail or add an error branch, and route on the status code. Errors come back as a sentence that names the problem, so you can pass it straight to a Slack or email alert:

None of these is billed: a request that fails costs nothing, so a retry never charges twice.

Workflow ideas

For the reverse direction — voice notes coming in — see transcribing Arabic voice notes.

What does it cost?

A 200-character reminder costs one cent. The full request reference, including other languages and saved voices, is in the API reference.