No vowels in the text
Written Hebrew omits niqqud, so the same consonant string can be several different words. Getting it right depends on modelling context, not just sound.
100+languages supported
MultilingualHebrew plus 100+ languages
Speaker IDwho spoke, and when
SRT · PDF · Wordexport formats
Why Hebrew is harder than the benchmarks suggest
Most speech models are trained on English-dominant data and evaluated on clean, read speech. Hebrew punishes that in four specific ways.
Root-and-pattern morphology
Prefixes and suffixes fuse onto roots (ו-, ב-, ל-, מ-, ש-, ה-). A model that segments badly produces text that is phonetically close and semantically wrong.
Constant code-switching
Real Israeli speech moves between Hebrew and English mid-sentence, especially in tech, medicine and law. Models that switch language wholesale lose the thread.
Fast, overlapping, informal
Meetings and calls in Hebrew run fast, overlap, and lean on slang. Clean-podcast accuracy does not predict how a model behaves here.
What you get back
Not a wall of text — a transcript you can actually work with.
Speaker separation
Each speaker is tracked and labelled with timestamps, so interviews and meetings stay readable.
Timestamps you can jump from
Click any line to hear the audio at that moment. Corrections take seconds instead of a re-listen.
Summaries and AI chat
Ask questions about the recording, or generate a summary of a long session, without leaving the transcript.
Subtitles and documents
Export SRT subtitles for video, or Word and PDF for reports, archives and legal files.
From recording to finished transcript
-
1
Upload or record
Bring an existing Hebrew audio or video file, or record directly in the app.
-
2
Transcribe
The transcript is built with speaker labels and timestamps.
-
3
Review and edit
Fix names and terms inline; the timing stays aligned as you edit.
-
4
Export
Send it out as SRT, Word, PDF or plain text, or keep working with it in the app.
Questions
How accurate is Hebrew transcription?
Accuracy depends heavily on the audio: a clean single-speaker recording behaves very differently from a noisy multi-speaker call. KolWrite publishes how it measures Hebrew accuracy, and on what material, on the Hebrew transcription accuracy page rather than quoting a single headline number that would not hold across conditions.
Does it handle Hebrew and English in the same recording?
Yes. Mixed Hebrew-English speech is normal in Israeli meetings and interviews, and is treated as ordinary input rather than a language switch that has to be declared up front.
Can it tell speakers apart?
Yes. Speakers are separated and labelled with timestamps, which is what makes interviews, panels and meeting recordings usable afterwards.
What can I export?
SRT subtitles for video, Word and PDF for documents, and plain text. Transcripts stay searchable inside the app as well.
Which devices does it run on?
macOS, Windows, Linux, iPhone, iPad, Android and the browser, with recording from Apple Watch and Wear OS.
Try it on your own Hebrew audio.
The only benchmark that matters is your own recording. Upload a file and compare the output yourself.
Start transcribing