Does speech to text understand Arabic dialects?

Partly. Speech to text tools handle Modern Standard Arabic best, and they make many more errors on dialectal speech ¹. How many more depends on the dialect, the tool and the recording, so no single accuracy figure describes how a tool will treat your speech. Test your own speech before you rely on it.

Modern Standard Arabic works best, and dialects fall behind it

Modern Standard Arabic is the formal written standard. A 2016 Interspeech paper says it is the official language of spoken and written Arabic media, while colloquial Arabic is the set of regional spoken varieties ². Many people speak a dialect at home and at work, so a speech tool is often hearing a variety it has seen less of.

A 2025 study says error rates rise sharply on dialectal and accented speech, compared with Modern Standard Arabic ¹. It names two causes: linguistic mismatch, and scarce labeled data.

On traditional Emirati Arabic, zero-shot Whisper models had 86 to 93 percent word error rates

Zero-shot means a model is used as released, with no training on the target speech. In a 2025 study, the authors built a small corpus of traditional Emirati Arabic and tested five speech models on it. Whisper Small had a word error rate of 93.06 percent, and Whisper Medium had 86.10 percent ³. At 86 percent, the tool makes roughly 86 errors for every 100 words spoken.

The corpus came from heritage broadcasts and literary sources, not everyday talk, and it covers one dialect. The same study found that training on the dialect helps. Fine-tuning improved both word and character error rates ⁴. Fine-tuning needs labeled recordings in your dialect, which most people do not have, so it is a tool for developers and researchers rather than a setting you can change.

Older tests show the same gap on Egyptian Arabic

The gap is old and well documented. On the 1996/97 NIST CallHome task for Egyptian Colloquial Arabic, the best word error rate reached 61 percent ⁵. That system is far older than today's models, so the number shows the size of the problem, not the error rate you will see now.

One figure can hide large differences between test sets

Accuracy depends on the test. Whisper large-v2 had a word error rate of 53.8 percent for Arabic on the Common Voice 9 test ⁶. Common Voice uses recordings from many volunteers in many conditions, so the figure reflects those speakers and recordings, and it is not a score for any one dialect. The question on the best Arabic speech to text API sets this figure beside a much lower one from another test set.

Pick the closest variety, then test it

Speech services list Arabic by locale, which combines a language with a country. Azure, for example, lists separate Arabic locales for Egypt (ar-EG) and Morocco (ar-MA) ⁷. A locale tells the service which variety to expect. It does not show how accurate the service is for each country.

Test the tool on your own speech with these steps:

1. Pick the Arabic entry closest to how you speak, even if it is not your country's. The platform lists are short, so our Arabic guide explains which entries each device offers.
2. Dictate two to three minutes of your normal speech, using the microphone you will really use.
3. Count the errors, then work out the rate using the method for calculating word error rate.
4. If the rate is too high, try another entry or another tool, or type the parts that matter most.

It depends on the speech you dictate

• Formal text, such as a letter or report written in standard Arabic, is closest to what the tools are built for.
• Mixed Arabic and English is harder. The code-switching entry explains why, and English words are where to proofread first.
• Casual talk is not what the figures above measure. The Emirati corpus came from broadcasts and literary sources, so it does not describe a phone call with a friend.
• Short-vowel marks are usually missing from dictated Arabic. The diacritics entry explains what they are and why they are often left out.

What to take from this

We did not find a published test that compares consumer dictation apps across several Arabic dialects. Treat any ranking that names a best tool for your dialect as a claim to test, not as a finding. For Modern Standard Arabic and the locales the platforms list, the built-in options in our Arabic guide are a reasonable place to start. If the text comes out in the wrong order around numbers or English, read why dictated Arabic text looks backwards.