Arabic IVR voice prompts with AI: a whole phone menu in one Gulf voice, in the right tone
Phone menus get patched for years by whoever is free to record, so the welcome sounds like one person and the holiday notice like another. With nutq-speak-3, every prompt comes from one voice, and each one gets the tone it needs.
nutq teamPublished 3 min read

Arabic IVR voice prompts are usually recorded once by a voice artist and then patched for years by whoever is free that day. The welcome sounds like one person, the Ramadan hours like another, and the outage message like a third. Text to speech fixes the consistency, but most Arabic text to speech reads a phone menu like a station announcement. nutq-speak-3 lets you generate every prompt in one Gulf voice and choose the tone line by line: warm for the welcome, plain for the options, and an apology on hold that sounds like it means it.
What prompts does an Arabic IVR need?
| Prompt | What callers need | Cue |
|---|---|---|
| Welcome | To feel received | [tender] |
| Main menu | Clarity, nothing else | none ([neutral]) |
| Queue and hold | Patience, a real apology | [tender] |
| Ramadan, Eid and holiday hours | Warmth, then the facts | [tender], then [neutral] |
| Outage or delay | Honesty without alarm | [sad], lightly, or none |
| Security warning | Firm and calm | none, with pauses |
| Promotion while on hold | A little energy | [excited], sparingly |
Two cues never belong in an IVR: [angry] and [scared]. A fraud warning read in a frightened voice makes callers anxious, not careful. Slow it down with pauses instead. Every cue, and how long each one lasts, is explained in Arabic text to speech with emotion.
How do you write prompts callers understand?
Keep the menu to four or five options and pause between them. Write numbers as words, so you control the form: «اضغط اثنين», not «اضغط 2». Brand and product names in English, such as Apple Pay, are read as English inside the Arabic.
A Gulf Arabic welcome and menu:
[Hind] [tender] هلا وغلا فيك، حياك الله في خدمة العملاء. [pause:500]
[Hind] [neutral] لخدمات البطاقات، اضغط واحد. [pause:400] للحسابات، اضغط اثنين. [pause:400] للتحدث مع موظف، اضغط صفر.
On hold:
[Hind] [tender] نعتذر على الانتظار، كل موظفينا مشغولين الحين. [pause:400] اتصالك يهمنا، وبنرد عليك في أقرب وقت.
A security notice, firm but calm:
[Hind] [neutral] تنبيه مهم: [pause:500] ما بنطلب منك أبدًا رقمك السري أو رمز التحقق. [pause:600] لا تعطيه لأي أحد.
Ramadan hours:
[Hind] [tender] مبارك عليكم الشهر. [pause:400] [neutral] أوقات الدوام في رمضان من الساعة تسع الصبح لين الساعة اثنين الظهر.
Government entities and banks often prefer Modern Standard Arabic for the menu itself. The same voices read fusha text, and Hassan is a Modern Standard voice if you want one built for it. Listen to the options in the voice library.
How do you generate and export the prompts?
Keep the prompts in a text file, one per line as file-name|text, and loop over them. The API returns a WAV; ffmpeg converts it to the 8 kHz mono that most phone systems take:
mkdir -p ivr
while IFS='|' read -r name text; do
curl -s -X POST https://nutq.dev/v1/speech \
-H "Authorization: Bearer $NUTQ_API_KEY" \
-H "Content-Type: application/json" \
-d "$(jq -n --arg t "$text" '{text: $t, voice: "hind", model: "nutq-speak-3"}')" \
--output "$name.wav"
ffmpeg -loglevel error -y -i "$name.wav" -ar 8000 -ac 1 -c:a pcm_mulaw "ivr/$name.wav"
done < prompts.txt
Use pcm_alaw instead of pcm_mulaw if your platform expects A-law, and keep the original WAV files as your masters. The full parameter list is in the API docs.
Will a prompt added next year match the rest?
Yes, if you keep two things fixed: the voice and the model. nutq-speak-3 gives the same reading for the same text in the same voice, and a new prompt in that voice sounds like the same person who recorded the rest. Name the model in each request, as the script above does, so a future change of default never changes how your menu sounds.
The same consistency makes fixes cheap. When opening hours change, edit one line in the file and regenerate one prompt, not the whole menu.
What about callers who want to talk, not press?
Prompts cover the fixed parts of a call. For a voice agent that listens and answers, nutq streams speech at 8000 or 16000 Hz for telephony and plugs into Vapi; see how to build a Gulf Arabic voice agent with Vapi. Many contact centres use both: recorded prompts for the menu, a live agent behind one of the options.
What does it cost?
$0.05 per 1,000 characters. A typical prompt is about 120 characters, so a 40-prompt menu is around 4,800 characters, or about 25 cents. Pauses are not billed. A phone menu is commercial use, which the monthly plans license. To try the voices first, type a line on the Arabic AI voice generator page, or start with the $5 of credit at sign-up.
Questions people ask
Can AI record Arabic IVR prompts?
Yes. You write each prompt as text and nutq-speak-3 speaks it in a Gulf Arabic voice, with the tone you mark: warm, plain or apologetic. Every prompt comes from the same voice, including ones you add months later.
Should an Arabic IVR be in Gulf dialect or Modern Standard Arabic?
Both are common. Government and banks often keep menus in Modern Standard Arabic; service brands increasingly greet callers in Gulf dialect. The voices read either, so you can produce both and offer a choice.
What audio format do phone systems need?
Most want mono WAV at 8 kHz, often µ-law or A-law; check your platform. nutq returns a WAV, and one ffmpeg command converts it.
If we add a prompt next year, will it match the old ones?
Yes, if you keep the same voice and pin the model. nutq-speak-3 gives the same reading for the same text and voice, so the voice sounds like one person across the whole menu.
What does a full IVR cost?
Speech is $0.05 per 1,000 characters. A 40-prompt menu of about 4,800 characters costs around 25 cents, so re-recording a changed menu costs less than a coffee.
Written by nutq team. nutq is an Arabic speech API built in the UAE: speech to text that keeps Gulf dialect, and Arabic text to speech with voice cloning.
اقرأ بالعربيةHear it on your own audio.
$5 of credit to start — about 33 hours of transcription. No card.
Keep reading
All articles
Text to speech3 min read
Free Arabic text to speech for Eid, Ramadan and wedding voice messages, with real warmth
In the Gulf, greetings are spoken: a voice note on WhatsApp beats a typed «عيدكم مبارك». nutq-speak-3 turns a written greeting into a warm Gulf voice, with the free credit at sign-up covering hundreds of them.

Text to speech4 min read
Arabic voice acting with AI for games and animation: characters that shout, tremble and laugh
A game character who sounds calm while a building collapses breaks the scene. nutq-speak-3 gives Arabic characters the range they need, angry, scared, surprised, laughing, from cues written into each line.

Text to speech4 min read
Arabic bedtime stories for kids, voiced by AI with real feeling
Children follow the voice before they follow the words. nutq-speak-3 reads a bedtime story the way a parent does: softer at the sad part, surprised at the twist, laughing at the silly one.