Arabic Text-to-Speech: Why It's Hard and How to Get It Right

Arabic text-to-speech is hard mainly because everyday Arabic is written without its short vowels, so the voice has to guess how each word is pronounced from the surrounding sentence. It also has to handle the split between Modern Standard Arabic and the spoken dialects, grammatical endings that change the sound of words, and numbers whose form depends on the noun they count. You can get natural Arabic audio today if you pick a voice that matches your audience, add vowel marks only where a word is ambiguous, and check the result line by line.

Why is Arabic text-to-speech harder than English?

English spelling is irregular, but it does write its vowels. Arabic mostly doesn't. Three things make the job hard:

1. Short vowels are usually missing. The voice has to work out whether three letters mean "he wrote", "it was written" or "books".
2. Written and spoken Arabic differ. Scripts are usually in Modern Standard Arabic, while daily conversation happens in dialects with different words and sounds.
3. Grammar changes word endings. Case endings, number agreement and connected-speech rules all affect what is actually said.

The right-to-left direction, which often gets the blame, is mostly a display issue. Unicode stores Arabic in reading order, so a text-to-speech system receives the letters in the order they are spoken. Direction only causes trouble in mixed text: an Arabic sentence containing an English brand name, a URL or a phone number can get scrambled when copied between apps. Check those lines before you generate.

What happens when Arabic is written without diacritics?

The Arabic alphabet writes consonants and long vowels. Short vowels and other pronunciation hints are small marks above or below the letters, called harakat or tashkeel: fatha (a), kasra (i), damma (u), sukun (no vowel), shadda (a doubled consonant) and tanween (the -an, -in, -un endings). The Quran, children's books and some poetry print them. Newspapers, websites, emails and almost every script you'd paste into a voice tool leave them out.

Take the three letters ك ت ب (k-t-b):

Written with vowels	Transliteration	Meaning
كَتَبَ	kataba	he wrote
كُتِبَ	kutiba	it was written
كُتُب	kutub	books

Without the marks, all three are simply كتب. The reader, human or machine, has to work out which one is meant from the rest of the sentence.

Two more examples:

• علم can be ʿilm (knowledge), ʿalam (flag), ʿalima (he knew) or ʿallama (he taught).
• ملك can be malik (king), malak (angel), mulk (sovereignty) or malaka (he owned).

Modern neural voices make this guess well in ordinary sentences, because they have learned from large amounts of text which reading fits which context. They slip in predictable places: headlines and short labels with no context, names, rare or technical words, poetry, and sentences where two readings are both grammatical.

A worked example

A slide title reads تعلم اللغة العربية. The first word could be taʿallum ("learning Arabic", a noun) or taʿallam ("learn Arabic!", a command). In a full paragraph the voice would probably choose correctly. On a four-word title it's a coin toss. Writing تعلُّم, with the shadda and damma, settles it.

How to fix a wrong vowel

Add diacritics yourself, but only on the word that went wrong. You don't need to vowel the whole script: one shadda or one fatha is often enough to tip the voice towards the right reading. For a word that appears many times, such as a product name or a person's name, add it to the pronunciation dictionary once instead of fixing every line. The pronunciation guide shows how.

Should Arabic voiceover use Modern Standard Arabic or a dialect?

Arabic has a formal standard, Modern Standard Arabic (MSA, or fuṣḥā), used for news, books, official documents and most written content across the Arab world. Alongside it, people speak regional dialects in daily life. Native speakers move between the two all the time, but they are different enough that a script written in one can't simply be read in the other.

English	MSA	Egyptian	Levantine	Gulf	Moroccan
now	al-ān	dilwaʾti	halla'	al-ḥīn	daba
I want	urīd	ʿāyiz	biddi	abgha	bghit

Sounds change too. The letter ق (qaf) is a deep "q" in MSA, but in Cairo, Beirut and Damascus it is usually a glottal stop, so qalb (heart) becomes ʾalb, and in much of the Gulf it is a "g": galb. The letter ج is "j" in MSA, "g" in Egyptian (gamīl, beautiful) and closer to the "s" in "measure" in Levantine speech.

A practical rule:

• Use MSA for news, documentaries, e-learning, corporate training, public information and anything aimed at several Arab countries at once.
• Use a dialect for advertising, social video, conversational characters and customer phone lines aimed at one country.

Write the script in the variety you want to hear. Choosing an Egyptian-sounding voice won't turn an MSA script into Egyptian Arabic, because the words themselves are different. Dialects also have no fixed spelling, so expect to listen more carefully and fix more words when you work in one.

How do case endings and pauses change Arabic pronunciation?

In formal MSA, nouns carry case endings: kitābun, kitāban, kitābin (a book, in three grammatical cases). At a pause, the final short vowel drops, so the word is said kitāb. Most newsreaders and narrators use a middle style: they pronounce endings inside a phrase where they help the listener and drop them at pauses. Text-to-speech voices generally follow this middle style, which is what most audiences expect.

Some other sound rules the voice has to apply:

• Sun letters. The article al- blends into certain consonants: الشمس is ash-shams (the sun), not al-shams, while القمر stays al-qamar (the moon).
• Linking alif. The vowel of al- disappears after a vowel: في البيت is fi l-bayt (in the house).
• Taa marbuta. The ending ة is -a on its own but -at before a following noun: مدرسة is madrasa, but مدرسة البنات is madrasat al-banāt (the girls' school).

Voices rarely break these regular rules; errors come from misreading the word itself.

If you need fully classical reading with every ending voiced, as in grammar lessons or poetry, provide fully vowelled text and check it closely. Quran recitation follows its own rules of tajwīd, and a human reciter remains the norm.

Why do numbers, dates and English words trip up Arabic voices?

Arabic number words agree with the noun in an unusual way: from three to ten, the number takes the opposite gender to the noun. Three books is ثلاثة كتب (thalāthat kutub), but three girls is ثلاث بنات (thalāth banāt). To read the digit "3" correctly, the voice has to know the gender of the noun after it.

Other common snags:

• Western digits (0–9) and Eastern Arabic digits (٠١٢٣٤٥٦٧٨٩) both appear, sometimes in the same document.
• Currency can be written "SAR 250", "250 ر.س" or in words.
• Dates can be Gregorian or Hijri, and the text often doesn't say which.
• English brand names and technical terms appear in Latin letters in the middle of Arabic sentences.

Where a number matters, such as a price in an ad or a year in a documentary, write it out in words. For English names, try both the Latin spelling and an Arabic-script spelling and keep whichever sounds right.

How do you make Arabic text-to-speech sound natural?

1. Decide on the variety. MSA or a specific dialect, based on who will listen.
2. Pick a voice. Listen to samples on the Arabic voices page and choose one whose accent and tone suit your audience.
3. Prepare the text. Use Arabic punctuation (، and ؟) so the voice pauses in the right places. Arabic prose often chains clauses with و (wa, "and"); for audio, break long chains into separate sentences.
4. Generate and listen line by line. In the Leelo studio each line is generated separately, so you can regenerate one line without redoing the rest and keep several takes to compare.
5. Fix vowels and names. Add diacritics to misread words and put recurring names in the pronunciation dictionary.
6. Direct the delivery. Adjust speed, pitch and pauses where the reading feels rushed or flat, as described in directing the delivery.
7. Get a native speaker to listen. A ten-minute review catches what no checklist will.
8. Export as MP3 or WAV.

Where is Arabic text-to-speech used?

• E-learning and training, where MSA narration keeps one course usable across many countries. See e-learning.
• Video, including dubbing existing English or French videos into Arabic.
• Phone menus, where prompts change often and a re-recording session for every change is slow. See phone menus.
• Articles and accessibility, giving readers an audio version of written content.

On cost, Leelo charges 1 credit per word in a standard voice and 1.5 credits in a premium voice. A 2,000-word course module is about 2,000 credits in a standard voice or 3,000 in a premium one. The free plan includes 1,000 credits once, enough to test a few voices on your own script, and paid plans are listed on the pricing page.

Frequently asked questions

Can text-to-speech read Arabic without diacritics?

Yes. Modern Arabic voices predict the missing short vowels from context and get ordinary sentences right most of the time. Mistakes concentrate in short headings, names, rare words and genuinely ambiguous sentences. Adding diacritics to just those words fixes them.

Does Arabic text-to-speech support dialects like Egyptian or Gulf Arabic?

It depends on the voice. Listen to the samples and choose one whose accent matches your audience, then write the script itself in that dialect. A voice can't convert MSA vocabulary into dialect vocabulary for you.

Can I use Arabic text-to-speech for Quran recitation?

It isn't a good fit. Recitation follows the detailed rules of tajwīd and carries religious expectations that synthetic voices aren't built to meet. Human reciters remain the standard.

How do I fix an Arabic name the voice keeps mispronouncing?

Add the name to the pronunciation dictionary with the fully vowelled spelling, or another spelling that produces the right sound. The fix then applies everywhere the name appears in the project, so you don't have to correct each line.

How much does it cost to dub a video into Arabic?

Dubbing uses 5,000 credits per minute of video, so a three-minute video takes 15,000 credits. If a job fails, the credits come back. Have a native speaker review the translated script before you publish.