Arabic AI voiceover for ads, reels and audiobooks: a production guide for Gulf brands

Gulf brands produce more video than ever — reels, TikToks, YouTube pre-rolls, app explainers — and almost all of it needs an Arabic voice. Booking a studio for every variation does not scale. This guide shows how agencies and creators produce Arabic AI voiceover that sounds local: casting, scripting, emotion, timing, and cost.

What can Arabic AI voiceover be used for?

How do you cast an Arabic AI voice?

Start from the audience. For an Emirati or Saudi audience, a Gulf voice; nutq has 16, male and female, most of them Emirati, plus Egyptian and MSA voices. Audition two or three on the same line before committing — the voice library lets you hear each one.

If the brand already has a voice — a founder, a presenter, an existing actor — clone it from six to eight seconds of clean speech and save it on the account. Every later generation uses it by name.

How do you write a script that sounds human?

Write in the dialect the listener speaks, then direct it with cues. A 20-second spot:

[excited] جديدنا وصل! [pause:300] قهوة مختصة، محمّصة في دبي، توصلك لين الباب.
[amused] وإذا ما عجبتك؟ [pause:400] [tender] نرجع لك فلوسك، ولا كلمة.
[neutral] اطلب الحين من التطبيق.

• One emotion per beat. The cue applies until the next one.
• Pauses are timing tools. [pause:300] lets a line land without slowing the whole read.
• Product names in English stay English. nutq-speak-3 reads English words inside Arabic as English.
• About 13 characters is one second — write to the length of the cut, not the other way round.

The full list of emotions and sounds, with examples, is in Arabic text to speech with emotion.

How do you produce a dialogue or a drama?

Open each line with the speaker's name, and the request returns one file:

[Sultan] [surprised] شو هذا؟ [pause:500]
[Maitha] [amused] هدية بسيطة [laughter] افتحها!

Each line is generated separately and joined, so it plays as two people reading lines rather than overlapping conversation. That suits ads and scripted drama; for long-form fiction, generate scene by scene and edit as you would a recorded session.

How do you handle long reads like audiobooks?

nutq-speak-3 is built for steadier long reads, so a narrator sounds the same in chapter ten as in chapter one. A practical workflow:

1. Split the book into scenes under 5,000 characters each (the per-request limit).
2. Keep the same voice and a neutral starting emotion across all scenes.
3. Add emotion only where the story turns: [sad] at a loss, [scared] in suspense.
4. Listen once at speed; regenerate only the lines that miss. Each take differs slightly, which is useful.

What does it cost?

Budget for retakes on top. Monthly plans include a commercial licence; check the current terms before running paid media. New accounts start with $5 of credit at sign-up.

Where to produce it

• In the browser: the Arabic text to speech page, or the studio, which also times Arabic captions to the same video.
• In code: one POST to /v1/speech; see Arabic text to speech with voice cloning.
• In automation: Arabic text to speech in n8n for batches of ads or daily content.