Hebrew speech-to-text API (עברית)

Transcribe Hebrew with one request

Set language to he and start a job. Poll GET /transcribe/{jobId} for the result, or pass a webhook_url.

curl https://app.kolwrite.com/transcribe \
  -H "x-api-key: $KOLWRITE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"audio_url": "https://example.com/interview-he.m4a", "model": "oracle", "language": "he", "include_diarization": true, "output_format": "json"}'

Translate Hebrew audio into English in the same job (Oracle models only):

curl https://app.kolwrite.com/transcribe \
  -H "x-api-key: $KOLWRITE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"audio_url": "https://example.com/interview-he.m4a", "model": "oracle_mini", "language": "he", "translate_to": "en", "output_format": "srt"}'

Hebrew at a glance

What to know about Hebrew transcription

• Hebrew is written right to left. The API returns plain Unicode text with no direction marks, so set dir="rtl" (or dir="auto") wherever you display it.
• Everyday Hebrew is written without vowel points (niqqud). Build search and matching for unpointed spelling, and strip niqqud from any reference text before you compare it with a transcript.
• Israeli speech mixes in English product names, acronyms and numbers. On mixed recordings, send language: "he" explicitly instead of relying on auto-detection, which decides on the dominant language.
• SRT files with Hebrew lines that end in punctuation or Latin words can render in the wrong order in players without bidirectional text support. Test your player, or add a right-to-left mark (U+200F) at the start of each cue when you post-process.

Where teams use Hebrew transcription

• Call and meeting recordings for Israeli companies, with speaker labels for who said what.
• Lectures and seminars at Israeli universities, published as Hebrew subtitles (srt).
• Journalism and podcast interviews, translated to English for international editors with translate_to: "en".
• Municipal, council and committee meeting minutes.

Choosing a model

Start with oracle (US$1.44 per hour) for production Hebrew audio. Use base (US$0.48) for clean, single-speaker recordings at volume, oracle_mini (US$0.96) for the cheapest translation, and oracle_pro or oracle_max for noisy or difficult audio. KolWrite does not publish a Hebrew accuracy figure for this API, so test a sample of your own recordings before committing. Full comparison: models and prices.

Questions and answers