Arabic is one written language and many spoken ones. A news bulletin, a university lecture read from notes or a Friday sermon is delivered in Modern Standard Arabic, fuṣḥā, which is the same from Morocco to Iraq. The moment the speaker relaxes, the language shifts to the dialect of the place: Egyptian, Levantine as spoken in Syria, Lebanon, Jordan and Palestine, Gulf, Iraqi, or the Maghrebi varieties of North Africa. These differ in everyday vocabulary, in verb forms and in sounds; the letter ج alone is pronounced three different ways across the region.
Recordings in Modern Standard Arabic come out best, because that is the form with a fixed spelling. Egyptian and Levantine speech, which are widely heard in film, television and online video, are also handled well at a normal pace. Dialects have no official spelling, so dialect words are written the way they are usually typed in messages and on social media, and two people would not always spell them the same way. Fast Maghrebi speech mixed with French, and strongly local rural speech, are where the transcript needs the most correction.
The transcript is written the way Arabic is printed in a newspaper or a book: right to left, with the letters joined, and without the small marks for short vowels, the tashkīl. That is normal for adult readers, who supply the vowels from context, but it means the text is not vocalised. If you need full diacritics, for a children's text or for teaching recitation, they have to be added afterwards by someone who knows the passage.
Numbers and Latin-script words sit inside right-to-left text, which some editors display in a confusing order when you paste. If a line looks scrambled after copying, set the paragraph direction to right-to-left in your document and it will fall into place. Proper names deserve a check: Arabic names have several accepted spellings, foreign names are written by sound, and the hamza and the tāʾ marbūṭa at the end of a word are easy places for small spelling slips.
Typical Arabic recordings here are long: religious lessons and lectures that run over an hour, university classes, recorded meetings, interviews for research or journalism, and podcasts and talk shows from YouTube. Voice notes forwarded from messaging apps arrive as OGG or M4A files and are accepted as they are. A video file needs no conversion either; the speech is taken from it and the picture is ignored.
Speakers often move between the standard language and dialect inside one sentence, and in business or technical talk they add English or French terms. Single foreign words are usually written in Latin letters or in their common Arabic spelling. Passages quoted from the Quran or from classical poetry are transcribed as ordinary speech, so check any quotation against the source before you publish it, since exact wording matters there.
Without an account you get 10 transcriptions of up to 3 hours each, enough for a long lesson or a full meeting in one upload. The free allowance is 5 hours in total across those 10 recordings, and a free account keeps the full transcripts of what you transcribed as a guest. AI chat with the transcript, automatic summaries and export to PDF or Word belong to the paid plans.
The speech models run on our own servers. The recording is not passed to a third-party speech API, which matters when the file is a private consultation, a client call or a family conversation that was never meant to leave the room.