Connect desktop dictation to your own server

A transcription server handles the audio, but you still need a way to use it while writing. This setup used OpenWhispr as the desktop app: a keyboard shortcut records speech, sends it to the chosen endpoint and inserts the returned text.

Connect the transcription service

In the version used for this build, OpenWhispr allowed a custom transcription address and API key. Those settings pointed to Run your own dictation server.

Test with a short, non-sensitive recording first. Check the returned text and the server logs before using the setup in another application.

Keep optional processing separate

Transcription turns speech into text. Optional rewriting or “reasoning” sends that text through another model and may use a different endpoint. Leave it disabled if you only want transcription, or check its destination separately.

Check the connection

The desktop app used a local HTTP address in this build. That connection is not encrypted by HTTP itself. Restrict access to an appropriate private network and check the app’s current requirements when setting it up.

If the connection fails, check the address, API key, network access and server response. If transcription works but optional rewriting does not, investigate that second endpoint separately. Do not assume one permission change will fix every connection error.

Review the privacy settings

Choosing a home server controls where the transcription request goes. It does not tell you what every other feature does. Check account syncing, optional cloud processing, logs and recording retention before using sensitive material.

On a work computer, use an approved setup. This project is for short dictation; it does not establish support for identifying speakers or producing meeting minutes.