Make a correction count.
Search and edit history, compare raw and cleaned text, re-run cleanup, and explicitly save a replacement from an edit.
For the ideas that move faster
than your fingers.
Send the draft to the HR team.
“Um, send the draft to the hr team.”
There’s more to dictation than recognising words.
Try the details that make them yours.
“Send the draft to the hr team.”
hr → HR
Names, brands, acronyms. Teach TawkType the spelling you want, or replace a phrase it keeps mishearing. Your vocabulary also applies after an optional rewrite.
Interactive illustration.Native Windows. A quiet status bar.
The controls you need, when you need them.
Switch modes with a key. See the active mode on the bar. All four work locally; choosing one never turns on cloud rewriting.
“um, check the api before we ship.”
Check the api before we ship.
Filler removed. A sentence starts with a capital.Illustrative example · Technical uses a saved “API” spelling.
Search and edit history, compare raw and cleaned text, re-run cleanup, and explicitly save a replacement from an edit.
If insertion fails, open the dictation box to edit, copy, or send text back when the destination supports it.
Hold to talk or use a separate toggle key. Esc cancels. A recording limit finishes a runaway dictation instead of discarding it.
No account. No subscription. No telemetry.
Speech recognition runs on your PC.
English and 24 other European languages. Parakeet runs on your CPU, so a dedicated GPU isn’t required.
Once a speech model is downloaded, local dictation works offline. History is stored locally when enabled; turn it off to stop saving dictations and delete the history file. Diagnostic logs record counts and timing, not your words.
Optional Claude rewriting can apply spoken corrections and format lists. It is off by default and uses your own Anthropic API key. When enabled, transcript text is sent to Anthropic; audio isn’t. API charges may apply.
Windows 10 or 11 and a microphone. A GPU is optional. The installer can fetch the .NET 8 Desktop Runtime if you need it.
Install TawkType, open Settings → Transcription, and choose a model to download. Models are not downloaded automatically. Select your microphone and language, then hold Right Ctrl, speak, and release.
TawkType targets the focused text field in editors, browsers, chat apps and other Windows programs. Compatibility varies. If it can’t insert text, it offers clipboard fallback and the dictation box.
Yes. Preferred spellings, replacements and spoken snippets can be imported and exported together as a vocabulary file. They work locally, with or without a Claude API key.
TawkType is not code-signed yet, so Windows SmartScreen may show a warning. The latest installer and release details are available in the project’s GitHub releases.
Get the Windows installer from GitHub releases.
Open Settings → Transcription to download one. Select your mic and language.
Click a text field. Hold Right Ctrl, speak, then release.
Your next good idea
is already on the tip of your tongue.