Unvocalized text
Written Hebrew leaves out vowel marks (niqqud), so one string of consonants can spell several different words. The model has to understand context to pick the right one, because sound alone isn't enough.
Hebrew transcription
Hebrew transcription built for how Israelis actually speak.
Upload a Hebrew audio or video file to get structured text you can read, edit, search, summarize, and export.
Live demo
Runs in your browser.
KolWrite works in 100+ languages, and here we look at what makes everyday Israeli speech so tough to transcribe.
- 100+languages supported
- MultilingualHebrew plus 100+ languages
- Speaker IDwho spoke and when
- SRT · PDF · Wordexport formats
Why Hebrew is harder than benchmarks make it look
Most speech models train mostly on English and get tested on clean, read-aloud audio. Hebrew trips those models up in four ways.
Root-and-pattern morphology
Hebrew attaches prefixes and suffixes directly to word roots (ו-, ב-, ל-, מ-, ש-, ה-). If a model splits the word in the wrong place, the text might sound close phonetically while meaning something completely different.
Switching languages mid-sentence
People in Israel mix English into Hebrew all the time, especially in tech, medicine, and law. Models that expect one language at a time quickly lose track.
Fast, overlapping, informal speech
In Hebrew meetings and calls, people talk fast, lean on slang, and talk over each other. Accuracy on a clean podcast does not predict how a model will behave here.
What you get back
A structured transcript you can put to work right away, rather than a block of raw text.
Speaker separation
We track and label each speaker with timestamps so you can actually follow who said what in an interview or meeting.
Clickable timestamps
Click any line to play the audio from that exact spot. You can fix a word in seconds without scrubbing through the whole recording.
Summaries and AI chat
Summarize a long session or ask questions about the recording right inside the transcript view.
Subtitles and documents
Export SRT files for video subtitles, or download Word and PDF files for reports, archives, and legal work.
From recording to finished transcript
Upload or record
Upload a Hebrew audio or video file you already have, or record straight in the app.
Transcribe
The system generates a transcript that includes timestamps and speaker labels.
Review and edit
Correct names and terms right in the text. The audio timing stays synced while you edit.
Export
Download the transcript as SRT, Word, PDF, or plain text, or keep working on it in the app.
Questions
How accurate is Hebrew transcription?
The audio quality makes a big difference. A clear recording of one person gives very different results from a noisy call with several speakers. Rather than quoting one headline figure that wouldn't hold up across every setting, KolWrite explains how it measures Hebrew accuracy and what material it tests on the Hebrew transcription accuracy page.
Does it handle Hebrew and English in the same recording?
Yes. Mixing Hebrew and English is standard in Israeli meetings and interviews. The system handles that mix as regular speech, so there is no need to declare a language switch ahead of time.
Can it tell speakers apart?
Yes. It separates speakers and adds timestamps to each turn, which makes recordings of interviews, panels, and meetings much easier to work with afterward.
What can I export?
You can export SRT subtitles for video, Word and PDF files for documents, or plain text. Your transcripts also remain searchable inside the app.
Which devices does it run on?
KolWrite is publicly available on macOS, Windows, Linux, iPhone, iPad, Android, and on the web, plus you can record from an Apple Watch. A Wear OS version has not been released.
Try it on your own Hebrew audio.
The real test is how it handles your own audio. Upload a recording and check the results for yourself.