Hebrew audio and video to text

Accents and spoken Hebrew

Everyday Israeli Hebrew is quite different from the written register: speakers drop formal forms, use slang and swap in words from English, Arabic and Russian. Accents vary too, including Mizrahi, Ashkenazi, Russian-accented and Arabic-accented speech. Clear, slow, single-speaker audio is transcribed far better than a noisy table of people talking over each other.

Script, vowels and right-to-left text

Hebrew is written without vowel marks, and so is the transcript. One word such as ספר can be sefer (book), sapar (barber) or safar (he counted), and context decides. Prefixes such as ב, ה and ש attach to the next word. Latin-letter terms and numbers inside a Hebrew line can display in a surprising order in some editors; that is a display matter, not lost text.

What Hebrew speakers use it for

University lectures recorded on a phone or Zoom, meeting notes, interviews for research or journalism, podcasts, and the long WhatsApp voice notes that are so common in Israel. Tech and business talk is often Hebrew with English nouns in the middle, which is the hardest case, so proof-read those lines first.

Which model works best for Hebrew

Hebrew is a lower-resource language for Whisper-family models, and the small model is noticeably less reliable here than for the major languages. Use clear, single-speaker audio, keep clips short, and plan to edit a good deal. For the best result, self-hosting with large-v3 helps.

On tryscriptorium.com the no-account preview and regular accounts run the small model (the larger medium and large-v3 models run when you self-host Scriptorium). In an account, pick Small in the model menu rather than Auto: Auto prefers the fastest, least accurate model for long files or a busy queue.

Hebrew data in the Whisper family is much smaller than for English, and the small model shows it, mainly in rare words, names and English terms said inside Hebrew. For important recordings, keep each clip to one speaker, check every name, or self-host with large-v3.

Subtitles and captions in Hebrew

SRT and VTT files are saved as UTF-8, so Hebrew displays correctly in YouTube and VLC. Keep lines short, around 40 characters, with two lines at most. If a full stop or comma appears at the wrong end of a line in one particular player, that is that player's right-to-left handling, and the text itself is fine. Subtitles stay in the language of the transcript, Hebrew.

The uploader is set to Hebrew

The upload box on this page transcribes as Hebrew, so the recogniser does not have to guess the language from the first seconds. If the recording is in another language or switches languages, use Audio to text, which detects the language automatically. Automatic transcription is a draft and no accuracy figure is promised: check names, numbers and anything you will rely on.