[Feature request] Hebrew (he) support for voice dictation
Feature request: Hebrew (he) support for voice dictation
Problem
Voice dictation in Claude (the desktop app, including the Code tab, and /voice in Claude Code) doesn't support Hebrew. When I speak Hebrew, the result is either English text or nothing usable. Claude itself reads and writes Hebrew very well, so the gap is only in speech-to-text.
ChatGPT and Gemini both dictate Hebrew well. Good open models for Hebrew exist too, for example the ivrit.ai fine-tunes of Whisper large-v3 and large-v3-turbo (https://huggingface.co/ivrit-ai). I run them locally on a Mac, and the turbo model transcribes a 17-minute dictation in about 18 seconds.
Why it matters
I work with Claude mostly by dictating long, detailed Hebrew messages (5 to 15 minutes each) for academic writing. Since Claude has no Hebrew dictation, I use a third-party dictation app. It often truncates or garbles long recordings, so Claude silently receives only part of what I said. To work around it, I had to build my own pipeline that re-transcribes each recording locally with the ivrit.ai models and cleans up the text. Native Hebrew dictation would remove all of this.
Proposed solution
- Add Hebrew (
he) to the supported dictation languages, using the language setting proposed in #31724. - Even if Hebrew accuracy is somewhat lower than English at first, it would be far better than no support at all.
- Long dictations (10 minutes or more) should come through complete, not truncated.
Related: #31724 (language setting for voice mode), #82971 (dictation always transcribes in English), #84777 and #78728 (requests for other languages).