Turn Arabic speech into text in your browser: drop a recording or a video and the words come back in Arabic script, as a plain transcript or as .srt and .vtt subtitles. Arabic speech to text works best on Modern Standard Arabic, the register of news, lectures and formal speeches. Spoken dialect goes through the same single Arabic setting and comes back with more mistakes, partly because dialect words have no fixed spelling to aim for. Text normally arrives without short-vowel marks (harakat), the way everyday Arabic is written. Whisper runs on your own machine, so the audio is never uploaded.
How it works
- Drop an Arabic audio file or video. Arabic is already set as the spoken language, which matters: Whisper writes in the language it is told to expect, and there is no auto-detect.
- Pick Accurate (≈77 MB) for dialect, phone recordings or several voices; Fast (≈41 MB) is enough for a first pass on one clear speaker. For subtitles, set Output to Subtitles before you start.
- Press Transcribe. Play any line back to check it, correct names and dialect words in place, then download .txt for notes or .srt / .vtt for captions.
Questions
- Can it transcribe Egyptian, Levantine or Gulf Arabic?
- Yes, through the same Arabic setting: Whisper has one Arabic language, not one per dialect. Expect more errors than on Modern Standard Arabic, and dialect words spelled however the model guesses. North African Darija, which borrows heavily from French, is usually the hardest. Use the Accurate model, and fix names and dialect words in place before you download.
- Can I get Arabic subtitles (.srt) from a video?
- Yes. Drop the video, switch Output to Subtitles before pressing Transcribe, and every line gets its own start and end time. Download .srt or .vtt: both are plain UTF-8 text with the Arabic stored in reading order, so players and editors that support Arabic display it right to left. A plain-text run carries no per-line timing, so pick Subtitles first.
- What about Arabic mixed with French or English?
- Each run decodes everything as the one language selected. With Arabic set, French or English phrases tend to come back respelled in Arabic letters, translated or garbled. If a recording switches language for long stretches, download the Arabic transcript, change Language to French or English and press Transcribe again; the audio is already decoded, so the second pass starts straight away.
- Does my file get uploaded?
- No. The models and all processing run inside your browser on your own device. You can verify this yourself: open your browser’s DevTools, watch the Network tab, and process a file: there are no uploads. After the first visit most tools also work fully offline.
- How accurate is it?
- The models are OpenAI’s Whisper (tiny and base tiers). On clear speech they are very usable; heavy accents, crosstalk and noisy rooms will show errors. The Accurate tier is noticeably better on difficult audio.
- Which languages work?
- Fifteen, picked in the Language setting: English, Spanish, French, German, Italian, Portuguese, Dutch, Russian, Polish, Turkish, Japanese, Korean, Chinese, Arabic and Hindi. The decoder is conditioned on the one you pick, so it transcribes that language rather than approximating it in English — which also means picking the wrong one returns confident nonsense. The weights understand 99 languages; these are the ones we measured as good enough to offer.
- How long does it take?
- Transcription runs on your processor, so speed depends on the machine: expect around a quarter of the recording’s length, and less on the Fast model than on Accurate.