Arabic speech to text

Transcribe Arabic Audio to Text

Upload or record Arabic speech and get an editable Arabic transcript with playback and timestamps on this page. The editor lays the text out right to left, so you can correct names, hamza and numbers, then download TXT, DOCX, SRT or VTT without leaving the page.

  • Arabic selected by default
  • Right-to-left editor
  • Mixed Arabic and English
  • TXT, DOCX, SRT and VTT
Arabic is preselected — change it if the recording is mainly another language
My recordsCloud processing · up to 1 GB / 60 minutes. Final usage is verified by the server.

Sign in to create a saved task. If your file is unavailable after signing in, select it again.

Dashboard

New Transcription
0 min

How do you want to transcribe?

Upload audio or video
Estimated cost: 0 min

Free minutes are included. Upload a file or record audio to start.

Choose your file and settings before signing in; sign in to submit. Processing uploads media to cloud services and uses account credits. Check the estimate and your balance first — see plans and usage limits and our privacy policy.

Answer first

What does Arabic transcription produce?

Arabic speech to text turns spoken Arabic into written Arabic you can edit, search and export. On this page Arabic is preselected, the transcript keeps the original language, and a right-to-left editor places playback, timestamps and corrections beside the text. English is a separate workflow, not the result. Treat automatic Arabic as a draft to check against the audio before you rely on it.

What happens on this page

Language
Arabic (ar) selected by default and changeable before each submission.
Direction
The editor and read-only view lay Arabic out right to left; timestamps stay left to right.
Result
An editable Arabic transcript with playback and timestamps; no automatic translation.
Downloads
UTF-8 TXT, DOCX and JSON from the saved text; SRT and VTT from the saved cues.
No speech
Empty output, bracketed non-speech tags or a known subtitle-credit line fails the task instead of being saved.
History
Tasks join My records with the rest of the workspace and reopen on this page.
Before you rely on it

Where review is still needed

  • Personal names, places and organisations, including the hamza carrier and any dialect spelling.
  • Numbers, dates and clock times, which may return as Arabic-Indic digits, Western digits or words.
  • Overlapping speakers, fast conversation, strong background noise or music.
  • Words that are valid but wrong for the sentence when short vowels are missing.
Three steps

How It Works

  1. 1

    Add Arabic Audio

    Choose an audio or video file, or record with your microphone. Arabic (ar) is already selected; switch to Auto-detect only if most of the speech is not Arabic. Add names or product terms to the important-words setting when it is shown. A file selection alone does not start a job — sign in, check the credit estimate, then submit from this page.

  2. 2

    Review Right to Left

    Upload and processing status stay on this page, and a refresh reopens the same task. When it finishes, play the recording beside the transcript. The editor lays Arabic out right to left, so the first word of a line is on the right and timestamps stay left to right. Correct names, hamza, numbers and punctuation, then wait for the saved confirmation.

  3. 3

    Export the Corrected Text

    Download the saved transcript as UTF-8 TXT, DOCX or JSON, or review the timed cues and export SRT or VTT. DOCX marks right-to-left paragraphs so Word keeps the Arabic reading order. The Arabic text is the result of this page; it is not translated into English automatically.

Practice files

Arabic Audio Samples You Can Run

Two short original clips with the human-written script each voice read. Download one, upload it above, and compare your result line by line.

Modern Standard Arabic · synthetic narration · macOS “Majed” voice (ar_001)
Human reference transcript
مرحبًا، أنا سارة الخطيب، صحفية في عمّان. في الرابع عشر من مارس، قابلنا سبعة وثلاثين موظفًا حول مواعيد العمل الجديدة. الخلاصة؟ يفضّل أكثر من نصفهم البدء قبل الساعة السابعة والنصف صباحًا. شكرًا لاستماعكم.

Check the hamza in أنا and أكثر, the shadda in يفضّل, the ta marbuta in سارة and الساعة, the Arabic comma ، and question mark ؟, and whether the spoken numbers come back as digits or as words.

Arabic with one English sentence · synthetic narration · macOS “Majed” (ar_001) and “Samantha” (en_US) voices
Human reference transcript
هذا تسجيل تجريبي لاجتماع فريق المنتج. سنناقش خطة الإطلاق في الربع الثالث. First, we review the beta feedback from last week. ثانيًا، علينا تحديث لوحة التحكم قبل يوم الجمعة. وأخيرًا، لنؤكد تاريخ الإطلاق في الخامس عشر من أكتوبر.

Check that the English sentence stays in Latin script and is not transliterated into Arabic, that الإطلاق keeps its hamza, and that the date الخامس عشر من أكتوبر is not silently converted to 15/10 or reordered.

The scripts were written for this page and read by built-in macOS voices, so the audio is clean, short and synthetic. They show what to check, not how well any model scores on noisy field recordings or on a regional dialect. The reference is the script, not a transcription output. Download both reference transcripts.

Review guidance

Arabic transcription and RTL review

Arabic is written right to left, so the transcript has to be laid out that way before it can be read comfortably. This page sets the direction on the editor and the read-only view.

Illustrated Arabic interview transcript laid out right to left with highlighted names, hamza and numbers beside a review checklist

In a right-to-left editor the first word of a line sits on the right, the caret moves the way you read, and a Latin name or a number inside the sentence keeps its own order. Without that direction, mixed lines such as “تقرير Q3 النهائي” can look scrambled even though the underlying text is correct.

Copy a corrected line into your own document to confirm the selection travels in logical order. Then work through the groups below with the audio playing: they are the corrections that matter most in an Arabic transcript.

Illustration created for this page, not a customer result. The practice clips above are the runnable example.

  • الهمزة

    Hamza and alef forms

    Arabic writes the glottal stop with several carriers: أ, إ, آ, ؤ, ئ and the bare ء. Speech recognition can pick a valid but wrong carrier, and a spellchecker will not flag it. Compare each name and each verb with how it is written in your source document.

  • التشكيل

    Diacritics that change the word

    Automatic output is normally undotted, so كَتَبَ (he wrote) and كُتُب (books) can look identical without context. Read the whole sentence before judging a word, and add short vowels only where a reader would otherwise misread a term.

  • ٠١٢٣

    Digits, dates and times

    The same number can come back as Arabic-Indic digits (٠١٢٣), Western digits (0123) or spelled words. A date may be reordered to a day/month/year pattern. Pick one style for the document and confirm amounts, dates and clock times against the source rather than the transcript alone.

  • ؟ ،

    Arabic punctuation

    Arabic uses the question mark ؟ and the comma ،, which point the other way from their Latin counterparts, and it does not use a capital letter to open a sentence. Automatic punctuation can leave Latin ? or , in place. Fix them before you export, especially in quoted speech.

  • Latin

    Latin names inside Arabic text

    Brand names, job titles and technical terms are often spoken in English inside an Arabic sentence. Keep Arabic as the language when Arabic dominates, then correct those words in Latin script. If a recording switches language for long stretches, compare Arabic with Auto-detect on a short excerpt.

  • RTL

    Caret, selection and copy

    A proper right-to-left editor puts the caret where you expect it and copies the logical order of the text, not the visual order. Select a mixed line such as “تقرير Q3 النهائي”, copy it, and paste it into your document to confirm the numbers and Latin word survive the round trip.

Bilingual speech

Mixed Arabic and English speech

Arabic conversations borrow English words all the time. The transcript has to keep both scripts readable in the same line.

When Arabic dominates a recording, leave Arabic selected and treat the English words as a review item. Most of them are brand names, job titles or technical terms, and they may be respelled or written in Arabic letters. Read the sentence back and restore the spelling your document uses, including the capital letters that Arabic script has no concept of.

For a file that genuinely alternates language for long stretches, test both settings on the first minute. Arabic is usually better when the amount of English is small; Auto-detect can follow a real switch but can also change its mind mid-file. Whichever you choose, the saved transcript is the record you correct.

Authored mixed-language example
00:00.0
سنناقش خطة الإطلاق في الربع الثالث.
00:06.4
First, we review the beta feedback from last week.
00:12.1
ثانيًا، علينا تحديث لوحة التحكم قبل يوم الجمعة.

What to check in a bilingual line

Script
An English word stays in Latin letters; it is not transliterated into Arabic.
Case
Arabic has no upper or lower case, so product names need their capital letters restored.
Direction
One RTL paragraph keeps the Arabic and Latin runs in the spoken order.
Digits
A quantity inside an English phrase may return as Arabic-Indic digits; normalise them.
Set expectations

Language and dialect limitations

Arabic is one language with many spoken varieties. One setting cannot promise equal accuracy everywhere, and this page does not pretend otherwise.

Modern Standard Arabic (MSA) is the shared written form used in news, lectures and formal interviews. Everyday speech is usually a local dialect with its own vocabulary and pronunciation. This page has one Arabic setting and does not rank dialects by accuracy because it has not measured them: test a short excerpt of your own dialect and judge the result yourself.

Dialect also changes the text, not just the wording. A speaker may use a word that never appears in MSA dictionaries, or drop the case endings an MSA writer would add. Keep what the speaker said instead of normalising it to MSA, unless the document is specifically meant to be in formal Arabic. When a recording matters, ask an Arabic speaker familiar with the dialect to check the passages you will quote.

What each variety means for review

Modern Standard Arabic
The shared written form: formal speech, news, lectures and read text.
Gulf, Egyptian, Levantine
Everyday dialects with their own slang and names; no accuracy rating is claimed here.
Maghrebi Arabic
Everyday dialects with French borrowing and fast elision; no accuracy rating is claimed here.
Code-switching
Review every English term; word-for-word handling is not guaranteed.

No accuracy percentage is published here. A single measured figure would not transfer from one recording to another, and the honest summary is the one above: test a short excerpt of your own dialect before committing a long file.

Before you upload

Supported Files, Silence and Plan Limits

Upload common audio and video formats such as MP3, WAV, M4A, AAC, FLAC, OGG, OPUS, WebM, MP4 and MOV, or record in the browser. The file picker lists every accepted extension. Each file can be up to 1 GB and 60 minutes; split longer recordings. A renamed extension does not convert the audio.

Recordings are uploaded to cloud services for processing; this is not an offline converter. Sign in to submit, check the estimate against your balance, and see plans and usage limits and the privacy policy. Only the account that created a task can open or download it.

A recording with no audible speech can fail instead of producing text. When a result is empty, holds only bracketed non-speech tags such as [موسيقى], or holds only a known Whisper subtitle-credit line, the task fails with a no-speech message and its transcription credits are returned. This filter targets empty and known non-speech output; it is not a guarantee that every invented sentence is caught, so check the transcript against the audio before relying on it. Very quiet speech under loud noise can still be missed or misheard.

Commercial comparison

WhisperWeb vs Notta vs Sonix vs TurboScribe for Arabic

A transparency comparison checked September 20, 2026. Competitor rows say “Not verified here” wherever this review did not test the product; only WhisperWeb's own behaviour is stated as fact. Names appear for comparison only and are not linked.

Arabic workflowWhisperWebNottaSonixTurboScribe
Arabic transcription on the pageWhisperWebAvailable: Upload, record, review hereThe file or recording, the Arabic editor and the export buttons are on this one route; a refresh reopens the saved task.NottaNot verified: Not verified hereThis review did not test Notta's Arabic editor, quotas or export list. Confirm it on the vendor's own site before you buy.SonixNot verified: Not verified hereThis review did not test Sonix's Arabic editor or export list. Confirm it on the vendor's own site before you buy.TurboScribeNot verified: Not verified hereThis review did not test TurboScribe's Arabic output or export list. Confirm it on the vendor's own site before you buy.
Language of the resultWhisperWebAvailable: Arabic stays ArabicArabic is forced as the input language and translation is off, so the task returns an Arabic transcript, not an English one.NottaNot verified: Not verified hereThis review did not test whether Notta keeps Arabic as the result or translates it. Confirm it on the vendor's own site before you buy.SonixNot verified: Not verified hereThis review did not test whether Sonix keeps Arabic as the result or translates it. Confirm it on the vendor's own site before you buy.TurboScribeNot verified: Not verified hereThis review did not test whether TurboScribe keeps Arabic as the result or translates it. Confirm it on the vendor's own site before you buy.
Right-to-left handlingWhisperWebAvailable: RTL editor and DOCXThe transcript editor and read-only view lay Arabic out right to left, and DOCX export marks Arabic paragraphs as right-to-left for Word.NottaNot verified: Not verified hereThis review did not test right-to-left caret behaviour or exported paragraph direction. Confirm it on the vendor's own site before you buy.SonixNot verified: Not verified hereThis review did not test right-to-left caret behaviour or exported paragraph direction. Confirm it on the vendor's own site before you buy.TurboScribeNot verified: Not verified hereThis review did not test right-to-left caret behaviour or exported paragraph direction. Confirm it on the vendor's own site before you buy.
Download formatsWhisperWebAvailable: TXT, DOCX, SRT, VTT, JSONUTF-8 throughout, keeping Arabic script. No PDF on this page: the shared PDF writer cannot render Arabic, so a PDF would be unreadable.NottaNot verified: Not verified hereThis review did not test Notta's current export formats or plan gating. Confirm it on the vendor's own site before you buy.SonixNot verified: Not verified hereThis review did not test Sonix's current export formats. Confirm it on the vendor's own site before you buy.TurboScribeNot verified: Not verified hereThis review did not test TurboScribe's current export formats. Confirm it on the vendor's own site before you buy.
What you can check before payingWhisperWebAvailable: Estimate and limits on the pageThe page shows the credit estimate before submit, states the 1 GB / 60 minute cap, the no-PDF decision and the dialect caveat, and lists the accepted extensions.NottaNot verified: Not verified hereThis review did not test Notta's current pricing or quota disclosure. Confirm it on the vendor's own site before you buy.SonixNot verified: Not verified hereThis review did not test Sonix's current pricing or human-service disclosure. Confirm it on the vendor's own site before you buy.TurboScribeNot verified: Not verified hereThis review did not test TurboScribe's current free allowance or price tiers. Confirm it on the vendor's own site before you buy.

Checked September 20, 2026. WhisperWeb's column describes this page's own implementation. Competitor columns are marked “Not verified here” because no competitor was tested for this page; they are not linked. No accuracy scores, ratings or price rankings are implied.

The honest advantage on this page is disclosure: the language, the right-to-left handling, the file cap, the credit estimate and the missing-PDF decision are all stated before you pay. A service with a human-transcription option may fit work that needs a professional proofreader, and a tool with a free daily allowance may fit occasional files. Because this review did not test the competitors, test the same short Arabic clip in each product yourself before committing a long recording.

Direct answers

Frequently Asked Questions

Will the result stay in Arabic?

Yes. Arabic is the input language and this page does not translate while it transcribes, so the saved transcript is written in Arabic script. If you need English, review the Arabic first and then use a page that is built for translation, such as the Audio to English Text tool, so the original Arabic stays in the task.

Are all dialects equally supported?

No. There is one Arabic setting rather than a separate option per country, and no dialect accuracy rate is published here because none was measured. Modern Standard Arabic, the written form used in news and formal speech, is the shared reference, while everyday dialects may use different vocabulary and pronunciation. Test a short excerpt of your own recording before a long job.

Can I export right-to-left text?

Yes. TXT, DOCX, JSON, SRT and VTT are UTF-8 and keep Arabic script. The DOCX export marks right-to-left paragraphs so Word keeps the reading order. PDF is deliberately not offered on this page: the shared PDF writer replaces Arabic characters, so a PDF download would be unreadable. Use DOCX and save it as a PDF from your word processor if you need one.

Does it transcribe Arabic and English in the same recording?

It can, and you should review the English words. Keep Arabic selected when most of the speech is Arabic; Latin brand names and phrases may be respelled. Check that an English sentence stays in Latin script instead of being transliterated into Arabic. For a recording that alternates language for long stretches, compare Arabic with Auto-detect on a short excerpt and keep the setting that needs fewer corrections.

What happens if my recording is silent or only music and noise?

A transcript is not saved for empty or known non-speech output. Speech models can invent text on silence, so when a result contains no letters or digits, only non-speech tags such as [موسيقى], or only a known subtitle-credit line, the task fails with a no-speech message and its transcription credits are returned. This is a filter for empty and known-pattern output rather than a guarantee that every hallucination is caught, so check the saved text against the audio.

Is Arabic transcription free or entirely local?

Processing runs in the cloud and uses account credits: up to 30 seconds costs 0.5 credit and longer files use one credit per started minute. Any free allowance depends on your account and plan. This is not an offline converter, so the selected media is uploaded for processing; see the pricing and privacy pages for the current limits. Selecting a file costs nothing until you submit.