Arabic · 22 varieties in our catalogue

Arabic text to speech, from Fusha to the dialects

Hear real Fusha clips made by sōtem V1. Fusha voices are available; dialect voices follow as native speakers are recorded, and Gulf voices come from sōtem Khaleeji, our own model, trained on recordings from paid, consenting native speakers.

Press play to hear it · Fusha · Man

Arabic · Fusha

مَرْحَبًا بِكُمْ فِي الْخِدْمَةِ الْمَصْرِفِيَّةِ الْهَاتِفِيَّةِ. أَنَا سُوتِم، مُسَاعِدُكُمُ الرَّقْمِيُّ. يُمْكِنُنِي مُسَاعَدَتُكُمْ فِي الِاسْتِعْلَامِ عَنِ الرَّصِيدِ، وَتَحْوِيلِ الْأَمْوَالِ، وَإِيقَافِ الْبِطَاقَةِ فَوْرًا عِنْدَ فِقْدَانِهَا.

0:00 / 0:13
How this clip was made

Nobody read this script aloud. The speech was generated by sōtem V1 on our own hardware, using a short reference recording from an openly licensed dataset.

Example voice made with a reference recording from the open ClArTTS dataset (CC BY 4.0).

Model
silma-tts by silma-aiLicence Apache-2.0 · revision 226dd7a
Reference voice
MBZUAI/ClArTTS (LibriVox public-domain recording), dataset revision 2f6e820abba8b5168b346c43479a324d8d37e35a, item ch_02_arabic_tts_dataset_192.wavLicence CC BY 4.0 · Attribution Kulkarni et al., ClArTTS, Interspeech 2023
Made
Generated on our own hardware (CPU) on 4 October 2026.
Independent check

An open speech-recognition model (Whisper large-v3-turbo, MIT licence, run locally) listened to the clip and heard:

مرحبا بكم في الخدمة المصرفية الهاتفية أنا سوت مساعدكم الرقمي يمكنني مساعدتكم في الاستعلام عن الرصيد وتحويل الأموال وإيقاف البطاقة فورا عند فقدانها

23 of 23 words matched · word error rate 4.3%

Custom text is available to pilot partners.

New voices, a man and a woman for each dialect from paid native speakers, come through sōtem Data Lab. Gulf voices: In training.

2 recordings
  • Fusha · Man

    Banking · مَرْحَبًا بِكُمْ فِي الْخِدْمَةِ الْمَصْرِفِيَّةِ الْهَاتِفِيَّةِ.…

    0:13
  • Fusha · Man

    Government service · نُفِيدُكُمْ بِأَنَّهُ تَمَّ اسْتِلَامُ طَلَبِ تَجْدِيدِ…

    0:12

Example voice made with a reference recording from the open ClArTTS dataset (CC BY 4.0).

Dialects

Every Arabic variety, with its honest status

Voices need a model trained for that dialect and licensed for commercial use. Where none exists, we say so and record our own.

AvailableIn developmentIn training

Gulf (Khaleeji)

  • Saudi Arabic · Najdi
    Text to speech: In trainingSpeech to text: Available
  • Saudi Arabic · Hijazi
    Text to speech: In trainingSpeech to text: Available
  • Saudi Arabic · Eastern Province
    Text to speech: In trainingSpeech to text: Available
  • Saudi Arabic · Southern
    Text to speech: In trainingSpeech to text: Available
  • Emirati Arabic
    Text to speech: In trainingSpeech to text: Available
  • Kuwaiti Arabic
    Text to speech: In trainingSpeech to text: Available
  • Qatari Arabic
    Text to speech: In trainingSpeech to text: Available
  • Bahraini Arabic
    Text to speech: In trainingSpeech to text: Available
  • Omani Arabic
    Text to speech: In trainingSpeech to text: Available

Iraq

  • Iraqi Arabic
    Text to speech: In developmentSpeech to text: Available (general Arabic)

Levant

  • Jordanian Arabic
    Text to speech: In developmentSpeech to text: Available
  • Palestinian Arabic
    Text to speech: In developmentSpeech to text: Available
  • Syrian Arabic
    Text to speech: In developmentSpeech to text: Available
  • Lebanese Arabic
    Text to speech: In developmentSpeech to text: Available

Egypt & Sudan

  • Egyptian Arabic
    Text to speech: In developmentSpeech to text: Available
  • Sudanese Arabic
    Text to speech: In developmentSpeech to text: Available (general Arabic)

Maghreb

  • Moroccan Arabic (Darija)
    Text to speech: In developmentSpeech to text: Available
  • Algerian Arabic
    Text to speech: In developmentSpeech to text: Available
  • Tunisian Arabic
    Text to speech: In developmentSpeech to text: Available
  • Libyan Arabic
    Text to speech: In developmentSpeech to text: Available

Yemen

  • Yemeni Arabic
    Text to speech: In developmentSpeech to text: Available (general Arabic)

Modern Standard

  • Modern Standard Arabic
    Text to speech: AvailableSpeech to text: Available
Speak a Gulf dialect? Get paid to record it

How a pilot works

Three steps to Arabic voices on your line

  1. 01

    Tell us about your service

    Who calls you, in which dialects, and what they need. We agree the pilot scope and pilot pricing for your organisation.

  2. 02

    We set it up for you

    The engine runs with the Arabic voices and transcription your callers need, with no third-party AI.

  3. 03

    Go live with real callers

    Start with a small group, listen to every clip and read every transcript. Every clip the engine makes carries a watermark.

Use cases

Where Arabic voices do the most good

  • 01

    Banks

    Card, account and fraud lines that speak the customer's dialect, with every call transcribed for compliance and never sent to an outside AI provider.

  • 02

    Government

    Citizens get answers about services, documents and deadlines by phone, day and night, and a person when the case needs one.

  • 03

    Contact centres

    Transcripts and summaries for every call in Arabic and English, so agents stop typing notes and supervisors can search every call.

  • 04

    Media

    Narration, promos and dubbing in natural regional Arabic, with a watermark that proves which clips are synthetic.

  • 05

    Education

    Lessons and audiobooks read in clear Fusha or in the dialect students hear at home.

  • 06

    E-commerce

    Order updates, delivery calls and product questions answered by voice, in the shopper's own dialect.

Questions

Arabic, answered

Fusha or dialect?

Fusha today: sōtem V1 reads Modern Standard Arabic in natural voices. Dialect voices follow as native speakers are recorded, and Gulf voices come from sōtem Khaleeji, our own model, trained on recordings we pay native speakers for.

Do I need to add tashkeel (diacritics)?

No. sōtem V1 reads Arabic with or without tashkeel. Adding it helps with names and words that can be read more than one way.

When will Gulf voices be available?

Gulf voices come from sōtem Khaleeji, our own model, in training on consented recordings from native speakers. No open Gulf voice model is cleared for commercial use, so we build our own.

Can one voice switch between Arabic and English?

sōtem V1 speaks both Fusha and English in the same voice. We test your own mixed scripts in the pilot before you rely on them.

How is it priced?

Pilot pricing per organisation, agreed before the pilot starts.

Where does it run?

In Europe, with your data stored in the EU (Frankfurt). No third-party AI API is called, and voices and text are never sent to an outside AI provider.

More answers on the sōtem Voice page.

Arabic voices for your callers

Tell us which dialects your callers speak and we'll set up a pilot for you.

Just exploring? Create a sōtem account

Type of organisation

We only use these details to reply about your pilot. See our privacy policy.