How to Test Whether Dictation Really Works Offline
To test offline dictation, install the local speech model, open a local text editor, disconnect from the internet, and dictate a new sentence. Then repeat with AI cleanup enabled if you use it. This checks whether each step works without a connection under your current settings.
Follow the audio and the text separately
| Stage | What it handles | What to verify |
|---|---|---|
| Speech recognition | Microphone audio becomes text | The selected recognizer runs on your computer |
| Optional text cleanup | The transcript and any rewrite instructions become an edited result | The provider and endpoint are local if you need the rewrite to work offline |
| Insertion and storage | The output goes to a text field and may be kept by the app | The destination, its sync settings, and any local history or clipboard retention meet your needs |
One common setup uses a local speech model and an online language model. In that setup, the audio can stay on your computer while the transcript goes to the configured provider for cleanup. The destination is a separate choice. A synced document can upload text even when speech recognition and cleanup both run locally. Voconly scenario settings
Run a short disconnect test
- Prepare while online. Install the speech model you want to use and make sure your microphone and shortcut work. Download and start the local language model too if you plan to use one for cleanup.
- Use a fresh local document. Pick a file outside synced folders. Use made-up text. For example, say “Log test eight five two, second pass.” Change the number for each run so you can identify fresh output.
- Test raw dictation. Turn off the rewrite step. Disconnect Wi-Fi and any wired network connection. Dictate and verify that the new text reaches your editor.
- Test your normal mode. Enable the exact cleanup preset you typically use and choose a change you can see, such as formatting two spoken tasks as a list. Check whether the formatting happens or only raw text appears.
- Record the setup. Note the app version, speech model, cleanup provider, network state, target editor, and what appeared. Check history and clipboard settings separately if you use those features.
| Result | What it tells you |
|---|---|
| Fresh raw text appears offline | The tested microphone-to-editor workflow can complete without an internet connection. |
| Raw text works, cleanup errors or produces the same raw text | Check the rewrite provider, model status, and whether the app fell back to plain transcription. |
| Neither mode produces text | Reconnect and verify the microphone, shortcut, and model first. A missing model or permission can also prevent output. |
A disconnect test establishes offline behavior for the run you performed. To evaluate what a tool sends when connected, review the relevant provider and app documentation and inspect the network behavior separately if you need stronger evidence.
How does this apply to Voconly?
Voconly can run speech recognition locally after the speech model is ready. For an entirely local draft, use plain transcription or a configured local text model and choose a local destination. Voconly's shortcut guide also describes a mode for bypassing the language model. If you enable an online provider, treat its inputs as data shared with that provider. Voconly shortcut guide · Voconly source and setup
If you're still choosing a speech model or setting up the microphone, start with the offline voice typing guide. Once plain transcription works, the test above lets you add each extra step and see how it changes the result.
Want to Try Offline Voice Typing?
Voconly is a free, open-source desktop voice input tool for Windows and macOS. It runs speech recognition locally, supports multiple ASR models, and can optionally process text with local or online LLMs.