Does Whisper support Arabic?

Yes. Whisper supports Arabic, and its code for it is ar ¹. Support is not the same as good accuracy. On the FLEURS read-speech benchmark, the large-v2 model had a word error rate of 16.0 percent in Arabic ², about 3.8 times its English rate on the same test ³. On dialectal and accented speech, error rates rise sharply compared with formal Arabic ⁴, so check your own variety before you depend on it.

Arabic is in the language list, with one code for many varieties

Whisper's language table lists Arabic with the code ar ¹. The table has one entry for Arabic. It does not separate the dialects that people speak in daily life.

That matters because Arabic is not one spoken language in practice. Modern Standard Arabic is the official language of spoken and written Arabic media, while colloquial Arabic is the set of regional spoken varieties ⁵. Whisper's single code covers both, so the model has to handle the formal language and many dialects under one setting.

Accuracy depends on whether the speech is formal or dialectal

The FLEURS figure is a read-speech result, so it shows the formal end of the range. The large-v2 model's 16.0 percent is the Arabic score on that test ², which is about 3.8 times its English rate on the same test ³.

Everyday speech tells a different story. On the Common Voice 9 test, which uses recordings from volunteers on many microphones, the large-v2 model had a word error rate of 53.8 percent for Arabic ⁶. That test is not the same as FLEURS, so do not subtract one figure from the other. The point is that a single Arabic score depends heavily on the test. A 2025 study found that error rates for Arabic speech rise sharply on dialectal and accented speech, compared with Modern Standard Arabic ⁴. The two Whisper figures on their own cannot show that cause.

It depends on the dialect, the setting and the app

• Formal speech such as news reading, lectures read aloud or prepared statements is the case closest to the FLEURS result.
• Dialect speech from a specific region, such as Gulf, Egyptian or Levantine speakers, is where error rates are most likely to rise, so test with a recording from your own region.
• The app matters. Apps built on Whisper may offer only some Arabic options, and dictation tools from other companies list Arabic by country and region. Apple's Mac list, for example, includes Arabic (Saudi Arabia) and Arabic (United Arab Emirates) ⁷. Check which Arabic setting your app actually sends to the model.
• Mixed Arabic and English can be harder again, because the model has to switch languages within a sentence. Set Arabic as the language for Arabic recordings, and expect more errors on English words.

What to do if you need Arabic

1. Set the language to Arabic rather than leaving detection on. Automatic detection looks at only the first 30 seconds of audio ⁸, and it can miss the language if the recording opens with silence or another language.
2. Test on the variety you speak. Dictate or transcribe two or three minutes of real speech, including a dialect or accent if you use one, and count the wrong words.
3. Use the largest model your app offers, and check that the app does not default to a smaller model.
4. For transcription, compare on the same recording. If you have a choice of tools, run the same minute of audio through each one and compare the errors rather than trusting a single published score.

The question on Arabic dialects and speech to text goes deeper on the varieties, and the question on transcribing Arabic or Hebrew audio covers the steps. The question on how good Whisper is at Japanese and the question on Whisper and Korean cover two other languages. For the wider picture, see the guide to Whisper's languages and accuracy and the overview of languages and accents.