Free, open-source dictation
Korean
voice typing for Windows.
한국어
Dictate Korean notes, messages and document drafts on Windows. With the shared All languages model installed, speech recognition stays on your PC.
Pin Korean, language code ko, for direct multilingual recognition. Pinning Korean together with English uses the same model with language detection, rather than downloading a second model for English.
Requires the shared All languages speech model (Whisper large-v3-turbo, approximately 1,621.7 MB). Download once in setup or Models, then reuse it offline.
How to set up Korean dictation
- Follow the current download and source setup instructions. Saydraft is a development preview for Windows 10 and 11.
- In Hub → Settings → Models, review the required model’s size and license. Choose Download / Resume if it is missing; Pause is available.
- Open Languages, enable automatic model choice, and select Korean. Save your settings. A single language pin uses code
ko. - Put the cursor in an editable app field, hold Ctrl + Win, speak, then release. Review the transcript before sending it.
Language selection starts no transfer. If automatic routing is off, the model selected on Models remains in use; select a compatible model there. For more detail, see the setup guide.
Use your voice in everyday Windows apps
Use Korean voice typing in a chat field, an email draft or your text editor. Review Hangul spacing, names and formal endings in the transcript. The dictionary lets you add names and specialist terms; Raw mode is available when you want to keep the recognized wording.
Speech recognition and text cleanup run locally. There is no account or cloud transcription fallback. With compatible models installed, you can dictate without an internet connection. Optional AI cleanup has its own model requirements; rules-only cleanup needs no text model.
Can I dictate Korean and English in the same workflow?
You can pin up to three languages. With Korean and English pinned, the router chooses All languages and lets that model detect the language. Recognition of short or mixed speech can vary, so review names and English terms in the result.
What does language support mean here?
This page describes speech-model coverage and app configuration. It does not claim a measured accuracy score for Korean. Accent, microphone, background noise and mixed speech affect results. The language choice does not translate the app interface or add Translate-mode output targets.