Docs
English dictation on Windows, offline
English is the language every speech model was trained on most, so the question is not whether HoldToType understands you but which of its nine models to pick: live text, the lowest latency, the widest reach, or the lightest one for an old laptop.
Updated · Vitalii Yemets
Which model to pick
The full list, with sizes and what each one is good at, is on the models page.
- Nemotron 3.5 shows the words on the plate while you are still speaking and punctuates by itself. The default for everyday dictation.
- Parakeet TDT 0.6B v3 turns a sentence into text in well under a second on a CPU, with punctuation, and covers 24 more European languages if you switch.
- Whisper Medium q5 is the safe general model: 99 languages and translation, at the cost of speed and about 1.5 GB of memory.
- Canary 180M is small, fast, and translates between English, German, Spanish and French without Whisper.
- Whisper Base or Small keep an old laptop comfortable; Qwen3 ASR is the most accurate and the heaviest.
Punctuation, names and your own words
Nemotron, Parakeet and Canary punctuate on their own; Whisper does it partly, and post-processing finishes the job. Names and terms that models spell wrong go into the dictionary: the spoken form on the left, the written one on the right. Replacements expand a short phrase into a long one, and voice commands such as new line do what they say.
Translation and other languages
Speak English and get the text in German, Spanish, French, Italian, Polish or Ukrainian, with Whisper doing the work on your disk. The other way round works too: one model per language, and the program switches when you change the language, so a French colleague can dictate on the same computer with Parakeet while you keep Nemotron.
On a weak machine
Everything runs on the processor; no graphics card is needed. On an old laptop start with Canary 180M or Whisper Base, both under 200 MB, and move up when you see how long a sentence takes. Live text with Nemotron needs a little more headroom than the others.
How to start
- Install from the download section; the wizard downloads a model for English.
- Put the caret in any text field, hold Ctrl+Win, say a sentence, let go.
- Change the keys under Shortcuts, or switch to toggle mode in Controls & behaviour.
- Add the names it gets wrong to the dictionary.
Questions
Does it understand accents?
The models were trained on a wide range of English, and Whisper and Qwen3 in particular are strong on accents. If a model struggles, try another: switching takes a click.
Is it as good as Windows voice typing?
On accuracy it is close, and it wins on names you can teach it and on privacy: Win+H sends your speech to Microsoft, HoldToType keeps it on the computer.
Does it need the internet?
No. Only the model download and the update check go online, when you press the button.
Dictate in English, offline
A minute to install, no account. Hold Ctrl+Win and say a sentence.