Transcribing Arabic is harder than transcribing English, and anyone who has tried knows why. People write Modern Standard Arabic but speak in dialects. A podcast host from Cairo, a voice note from Riyadh and a news bulletin all sound very different, and speech recognition tends to do better with formal Arabic than with everyday dialect speech.
You can still get good transcripts. This guide covers how to set up a transcription for Arabic, what to expect from dialects, how to deal with noisy audio, and how to clean up the text and turn it into subtitles that read correctly from right to left.
Know what you're transcribing
Before you start, it helps to name what's in the recording:
- Modern Standard Arabic (MSA, الفصحى): news, lectures, formal speeches and most written content. Speech recognition usually handles it best.
- A dialect (العامية): Egyptian, Gulf, Levantine, Iraqi, Maghrebi, Sudanese and others. Everyday conversation, vlogs, podcasts and voice notes are almost always in a dialect.
- A mix: many speakers move between MSA and their dialect, and between Arabic and English or French, sometimes in the same sentence.
The more dialect and mixing there is, the more editing you should plan for. That's not a flaw in your audio; it's simply the hardest case for any speech recognition model.
Choosing the right settings
Spokly's free audio and video to text tool runs in your browser, so your recording isn't uploaded. Two settings matter most for Arabic.
The spoken language
The language list includes Modern Standard Arabic and Egyptian, Gulf (Saudi, Emirati, Kuwaiti), Levantine, Iraqi, Maghrebi (Darija) and Sudanese Arabic.
It's worth knowing how this choice is used:
- For transcribing a file, the speech model has one setting for Arabic, so all the Arabic choices tell it the same thing: "this is Arabic". Choosing Arabic instead of "Auto-detect" is still important, especially for short clips, recordings that start with music, or speakers who mix in English words. Without it, the model may decide the audio is in another language.
- For voice typing, the dialect choice does change something: Spokly asks the browser for the matching regional setting, such as Arabic as spoken in Egypt for Egyptian. How much that helps depends on the browser's speech service.
Fast or Accurate
Choose Accurate for Arabic, especially for dialects. It's the larger model: the one-time download is bigger and transcription takes longer, but it usually handles dialect words, fast speech and mixed languages better than Fast.
Use Fast only for clear MSA, or to check quickly whether a file is worth transcribing in full. Our guide to free transcription without uploading explains the download sizes and what to expect the first time.
Dealing with dialects
Even with the right settings, dialect speech often comes out in one of these ways:
- Pulled toward MSA: the model may write a dialect word as its closest standard form. For example, Egyptian "عايز" might come out as "أريد", or a Gulf word might be replaced by a more common one.
- Spelled phonetically: dialect words have no fixed spelling, so the same word can be written in different ways within one transcript.
- Missed or merged words: short dialect words and quick particles can be dropped, or two words can be joined.
What helps:
- Decide on a spelling style before you edit. If you're transcribing Egyptian speech, decide whether you keep it in dialect (as spoken) or move it toward MSA. Mixing both styles in one text is the most confusing result for readers.
- Keep a short list of recurring words (names, places, brand names, expressions the speaker repeats) and fix them all with search and replace.
- Listen again where meaning matters. For quotes, numbers and names, check the audio rather than trusting the text.
Recording and preparing noisy audio
Arabic content is often recorded in cafés, cars and busy homes. Noise affects every language, but it hurts more when the model is already less sure about dialect words.
If you're recording
- Hold the phone or microphone close to the speaker's mouth.
- Turn off air conditioning or fans near the microphone if you can.
- Record each person on their own microphone for interviews, if possible.
- Avoid music under the speech. Add it later in editing instead.
If the recording already exists
- Use the original file, not a version that was sent through a messaging app and compressed again.
- Cut long music sections, intros and outros before transcribing; they can produce stray words.
- If one part is very noisy, transcribe it separately and plan to type that section by hand.
Cleaning up the transcript
A quick, organized edit is faster than fixing words at random. Work in passes:
- Names and terms. Fix people, places, brands and technical words first, using search and replace for repeated ones.
- Common letter variations. Speech recognition and fast typing both mix up similar letters. Look for ة and ه at the ends of words, ى and ي, and the forms of alif (ا، أ، إ، آ). Choose one consistent style.
- Punctuation. Add the Arabic comma (،) and question mark (؟), and break the text into sentences and paragraphs where the speaker changes topic.
- Foreign words. Decide whether English words stay in Latin letters or are written in Arabic letters, and keep it consistent.
- Numbers. Choose Arabic-Indic digits (١٢٣) or Western digits (123) for the whole text.
The transcript doesn't label speakers, so for interviews add the names yourself as you go.
Arabic subtitles and right-to-left text
You can turn the same recording into subtitles with the subtitle maker on the Spokly home page. Choose Arabic, make the subtitles, fix any cue text right on the page, then download SRT or VTT. Spokly saves these files as UTF-8, which is what platforms expect for Arabic.
Right-to-left text needs a little extra care:
- Check in the real player. Upload the file to your platform and watch a few cues before publishing. Different players handle right-to-left text slightly differently.
- Watch the punctuation. When a line ends with a full stop or question mark, some players show it on the wrong side. If you see this, simplify the punctuation at line ends, or test the other format (SRT or VTT) in that player.
- Mixed Arabic and English lines can appear in a confusing order. If a cue mixes scripts, try rephrasing it so the English word isn't at the start or end of the line.
- Keep lines short. Arabic subtitles are easier to read when each cue is one or two short lines.
For the upload itself, see how to add subtitles to a YouTube video.
A realistic workflow
Putting it together, a typical Arabic transcription looks like this:
- Trim music and silence from the recording.
- Transcribe with the language set to Arabic and Accurate selected.
- Read the transcript once, marking anything that needs a second listen.
- Fix names and terms with search and replace.
- Normalize spelling, punctuation and numbers.
- Make subtitles if you need them, and check them in the player.
Spokly's AI tools for cleaning up and rewriting transcripts, including Arabic dialects, are coming soon. Transcription and subtitles are free today.
FAQ
Which Arabic option should I choose for an Egyptian podcast?
Choose Arabic – Egyptian. For transcribing a file, all Arabic choices tell the model the same thing, so the important part is choosing Arabic rather than Auto-detect, and choosing Accurate. For voice typing, the Egyptian choice also selects Egypt's regional speech setting in the browser.
Why does my transcript look like formal Arabic when the speaker used dialect?
Speech recognition tends to do better with formal Arabic, so it often writes dialect words in a more standard form. Decide which style you want and edit the transcript toward it.
Can I transcribe a recording that mixes Arabic and English?
Yes, but expect to fix the English words. Set the language to Arabic if most of the speech is Arabic, and check names and English terms carefully afterwards.
Is my Arabic recording uploaded anywhere?
No. File transcription runs in your browser, on your device. Voice typing is different: in Chrome and Edge, the browser's speech service processes your voice.