Troubleshooting

Start with this quick check#

1. Confirm TawkType is running and note the bar’s status.
2. Open guided setup and confirm the microphone meter responds.
3. Confirm the selected model is downloaded and loaded.
4. Try a short dictation in the built-in practice box or a simple editor.
5. Compare that result with the app where the problem occurs.

A successful practice dictation narrows the problem: the microphone and recognizer have worked. If delivery fails elsewhere, keep the completed text and investigate the destination rather than immediately reinstalling everything.

The shortcut does nothing#

• Make sure the app has not been quit; the hotkey only works while it is running.
• Check the shortcut shown on the bar and in Settings → Activation.
• Hold it longer than the tap threshold. The default is 250 ms.
• Record the shortcut by pressing the physical keys, then Save.
• Try a different combination if another app intercepts it.

If you intentionally cleared the toggle shortcut, use the hold key. Pressing an unassigned toggle cannot start a recording.

The microphone meter does not move#

Check the selected input, its physical mute switch, Windows input level and desktop microphone permissions. A headset connected after setup may need to be selected again. Test the device in Windows’ input controls to distinguish a device problem from an app problem.

Setup requires an observed signal before it continues. Clicking Continue or typing in the practice box does not substitute for microphone input.

The model is missing, downloading or will not load#

Open Settings → Transcription and check the saved engine, language and model. A language change can select a different engine in Auto mode. Detect automatically routes to Whisper, so a Parakeet download alone is insufficient for that choice.

The download picker is separate from your active configuration. Save any engine or language change, reopen Transcription and select the matching model in the download picker. See the model download steps.

For a failed download, check internet access and disk space, cancel if needed and retry. If files are present but loading consistently fails, use the app’s model-management controls to remove the affected model and download it again. Close other memory-heavy apps and review your graphics driver if using Whisper’s GPU path.

The setup model step waits for initialization as well as download. Large files being present does not prove they can be loaded.

Recognition finishes but no text appears#

Read the status bar and open recovery. If the result was copied, click the desired field and paste. Check whether the target app runs as administrator, the control accepts text and you kept focus in place until delivery.

Try Auto or Paste in Settings → Activation. If a simple editor works but a custom app does not, use the dictation box for deliberate delivery. Inspect partially inserted text before retrying so it is not duplicated.

A word is changed or disappears#

Compare Heard and the final text in History. If Heard is wrong, check language and microphone quality, compare models or add a targeted replacement. If Heard is right but the final version is wrong, inspect the active mode, vocabulary and optional rewrite.

Filler removal can collide with legitimate lower-case words in some languages. Use Literal or turn that switch off. A preferred spelling fixes a matching phrase; a different recognized phrase needs an explicit replacement.

A snippet does not expand correctly#

• Save a rule in the form trigger => saved text.
• Say “insert” followed by the trigger; do not store “insert” twice.
• Inspect Heard to see whether the trigger was recognized.
• Use a literal backslash plus n for a line break in the settings editor.
• Remove overlapping rules that alter the expanded text.

An expanded snippet deliberately skips Claude. That is expected even if the rest of the dictation might otherwise have been rewritten.

Claude seems to do nothing#

Check all three requirements: the global rewrite switch is on, a key is available and the active mode allows the pass. Literal blocks it by default, and a snippet expansion skips it. Check the model identifier and network connection.

The current implementation falls back to local text on a provider error or timeout, which can look like an unchanged result. The default timeout is two seconds. You can lengthen it, but a larger delay is then possible. TawkType is a dictation cleaner, so dictated requests such as “write me a poem” are intended to remain text, not be carried out.

Settings will not save or history is empty#

For a disabled Save button, read the numeric validation message. Use the units shown in the label: the tap threshold and rewrite timeout are milliseconds, while the recording limit is minutes.

For an empty History window, check whether logging was enabled when the dictation happened, whether a search filter is active and whether retention removed older entries. The in-memory latest result is separate from persistent history. A new microphone recording cannot be reconstructed from a text-only history entry.

Report a useful issue#

Use the project issue tracker. Include the installed version, Windows version, engine/model/language, destination application, selected mode and insertion method. Describe the steps, the bar’s message and whether the same sentence works in the practice box.

For recognition or cleanup issues, use a harmless example of Heard and final text. Review logs and screenshots before attaching them so you do not expose transcripts, API keys, private names or unrelated files. The logs are in the data folder described in Privacy & your data.