Your first dictation

Open guided setup#

Setup opens automatically for a new configuration. To reopen it, right-click the TawkType tray icon and choose Set up TawkType…. The walkthrough saves each choice as you make it so the microphone, model and practice steps use your real settings.

You can go Back to revisit an earlier step. Skip setup leaves the walkthrough without claiming the microphone or model is ready. It does not keep reopening at every launch; return from the tray menu when you are ready.

1. Welcome#

Read the short introduction: hold a key, speak and release. Speech recognition runs on your PC. The optional Claude rewrite is the separate feature that can send transcript text to an external service. Continue to the microphone check.

2. Check your microphone#

1. Choose the microphone you intend to use, or keep (system default) to use the Windows default input.
2. Say a sentence at your normal speaking volume and watch the live meter.
3. Continue once TawkType has heard a level above the noise floor.

The message The meter has to move first means setup has not yet detected sufficient input. Check a hardware mute switch, input selection and Windows microphone permissions. Changing devices resets the check so an earlier device cannot stand in for the new one.

The setup meter uses the selected microphone while this step is visible. It stops when you leave the step or hide the setup window.

3. Choose a language and prepare a model#

Choose the language you will dictate. Auto engine selection uses Parakeet for its supported set of 25 languages and Whisper for the remaining choices. Detect automatically uses Whisper when the engine is set to Auto.

1. Review the selected model and download size.
2. Start the model download. Parakeet is approximately 640 MiB on disk; the default Whisper model is approximately 1.5 GiB. The download display may use rounded MB/GB values.
3. Wait for both downloading and loading to complete. A file on disk alone does not pass this step.
4. Continue when the model is reported ready.

Download the model first means the files are missing. Wait for the model to load means initialization has not completed. An already downloaded model can be loaded without fetching it again. You can cancel a download and retry later.

4. Record your shortcut#

Click the shortcut box and hold the key or combination you want. Do not type the name of the key. The default is Right Ctrl. Setup checks whether the combination is usable and reports known conflicts; a warning cannot cover every shortcut used by every other app.

Esc leaves a shortcut picker unchanged. Choose a combination you can comfortably hold while speaking, then continue. Optional toggle and mode-switch shortcuts can be configured later.

5. Dictate into the practice box#

1. Click inside the practice text box so it owns the caret.
2. Hold your chosen shortcut for longer than a quick accidental tap.
3. Say something simple, such as “This is my first TawkType dictation.”
4. Release the shortcut and wait for the completed text.

The practice step uses the real microphone and dictation pipeline. Typing into the box yourself does not prove speech recognition worked. If the step remains blocked, read its status, revisit the microphone/model steps and try again.

6. Choose your privacy preferences#

Decide whether to keep a local history of your dictations and whether optional cloud rewriting is enabled. History is plain text on your PC. Cloud rewriting still needs a configured Anthropic key; the permission setting alone does not supply one.

Local vocabulary and cleanup remain available with Claude off. If you turn history off, TawkType stops retaining the dictation log and attempts to remove its history file. A completed last result can still be available in memory for recovery while the app is running.

7. Finish and use another app#

The final page reports what setup actually observed: microphone input, a loaded model and a successful dictation, plus the privacy choices in effect. Finish setup, open an editor such as Notepad and try the same hold–speak–release sequence there.

If dictation succeeds in the practice box but fails in another app, the microphone and recognizer have already worked. Start with the destination field, insertion method and recovery options rather than reinstalling the model.