Offline Speech-to-Text and AI Cleanup
Set up private Windows dictation with local Whisper and S1-mini AI cleanup. Download models, choose a writing style, and work without cloud calls.
Speech-to-text and AI cleanup without the cloud
Use local Whisper transcription together with local transcript cleanup to turn speech into written English on your Windows PC. Local cleanup is available on every plan, including Free, and works in Local-Only mode.
Set up offline dictation
- Install Dikt and choose a local Whisper model. Download the model while connected to the internet. See the Whisper model comparison for the available sizes.
- Open Settings → Transcription → Cleanup → Local Transcript Cleanup and enable local transcript cleanup.
- Download the S1-mini cleanup model. This is a one-time 462 MB download, with roughly 600 MB of RAM needed while the model is active.
- Choose Semi-formal, Casual, Semi-casual, or Formal writing style, then save your settings.
- Enable Local-Only Mode in Settings → Advanced to block cloud calls. Once the models are downloaded, press your dictation hotkey in a text field and speak.
What local cleanup changes
S1-mini removes fillers and false starts, applies spoken self-corrections, and formats numbers, dates, and email addresses. For example, “the fifteenth, no wait, the sixteenth” becomes “the 16th”. Review the result before sending or publishing it.
Local cleanup supports English in this release. Whisper transcription supports other languages with a multilingual model. If you also configure cloud AI cleanup outside Local-Only mode, it runs after local cleanup and sends the resulting text to that provider.
If setup does not finish
Finish model downloads before disconnecting. A cancelled cleanup download can be resumed in the same panel. You can disable cleanup or delete its model there to reclaim resources. For recording or text insertion problems, see troubleshooting.