Documentation

For Voyce 1.0.1 on Windows. Last updated 24 August 2026.

Installing

Download the installer and run it. It installs for the current user only, so it never asks for an administrator password and never touches anyone else's account on the machine.

First run

Voyce is about 211 MB to download because the fast speech model is inside it. There is nothing to fetch before you start.

Install it, activate your licence, and dictate. Holding the key works on the first run, offline, with no model download and no account. That is why the installer is the size it is: roughly 140 MB of it is the speech model itself.

The app works with no connection at all. You can take the laptop off wifi and it keeps going. The only step that needs the internet is activating your licence once.

There is a second, more accurate model (about 464 MB) under Settings if you want it. It is optional, it is slower, and Voyce never downloads it on its own or starts a download while you are dictating.

Dictating

Put the cursor where you want words to appear, in any application. Hold Right Ctrl, say the sentence, and let go. The text types itself into that window.

There is no box to copy out of and nothing to paste. Voyce types where you were already typing, which is why it works in Word, Slack, a browser, a terminal or a game chat without knowing anything about them.

Hold or toggle

By default you hold the key for as long as you are speaking. If you would rather press once to start and once to stop, switch Mode to toggle in the panel. Toggle suits long dictation; hold suits short bursts and is harder to leave running by accident.

Changing the push-to-talk key

Open the panel, find Push-to-talk key and click Set key. The next key you press becomes the new one. If you change your mind, press Esc or click away: the capture cancels itself cleanly and the old key keeps working.

Pick a key you do not otherwise use. Right Ctrl is the default because most people never press it deliberately.

Every setting in the panel

Setting	What it does
Push-to-talk key	The key you hold to dictate.
Mode	Hold the key while speaking, or press once to start and once to stop.
Quality	Fast or accurate. Accurate uses a larger model and is a Pro feature; free accounts stay on fast.
Microphone	Which input device to record from. Leave on the system default unless you have a reason.
Start with Windows	Voyce is running and ready at every login.
Add a space after	Puts a space after each dictation, so consecutive sentences do not run together.
Sound cues	A short beep when recording starts and stops.
On-screen status pill	A small floating indicator showing when Voyce is listening.
Send error reports	Off by default. See what leaves your PC.
Stop automatically	Pro. Ends the recording when you stop talking, so you can let go of the key early.
Live preview	Pro. Shows words as you speak rather than all at once at the end.
AI formatting	Pro. Tidies punctuation and capitalisation. It rewrites the transcript, so leave it off if you want exactly what you said.

Teaching Voyce your own words

Names, jargon and product names are what speech recognition gets wrong, and they are exactly the words you use most. Voyce learns them from you.

In the panel there is a box where you can dictate to test. When it mishears something, correct it there. Two things happen: the word joins a glossary that biases the next decode toward it, and a repair rule fixes that same mistake immediately next time.

An ordinary English word needs to be corrected twice before it is trusted. One slip of the finger cannot teach Voyce to rewrite a common word everywhere.

This is deliberately the only place teaching happens. Voyce does not watch what you type in other applications to learn from corrections you make there, because that would mean reading your work in Word and your messages in Slack.

What Pro unlocks

Free is 15 minutes of dictation a day, forever, on the fast model, with no account and no card. It is not a trial and it does not expire.

Pro removes the daily limit and adds the accurate model, automatic stop on silence, live preview as you speak, and AI formatting. It is £8 a month or £149 once.

Licences and devices

A licence is a key you paste into the panel and activate. There is no account to create and no password to remember.

• Lifetime keys are verified entirely offline, against a public key built into the app. They never contact a server, so a lifetime licence keeps working whatever happens to us.
• Monthly keys check in periodically to confirm the subscription is still active. The check sends the key and nothing else.
• If that check cannot reach us, a paid subscription keeps working for 14 days offline. Our servers being unreachable is our problem, not yours.

What leaves your PC

Your audio and your transcripts do not. There is no upload step in the app, and no transcription service behind it. What you dictate exists on your machine and nowhere else.

Shared vocabulary. When you correct a word, that pair of words is sent to a shared list so your colleagues benefit from it, and your app receives theirs. Text only: no audio, no transcripts, no name, no email and no licence key. A correction is withheld until three separate machines have made the same one independently, so a one-off correction, such as the spelling of a single customer's name, is never passed on to anybody. Your own corrections always take priority over the shared list. Turn it off with Share vocabulary with my team in Settings.

Two things can travel, and both are narrow:

• A monthly licence check, which sends your licence key so we can confirm the subscription is active. Lifetime licences never do this.
• An error report, only if you switch it on. It is off unless you turn it on. What it contains is the app version, your Windows version, which feature failed, and the error message, each capped in length. Never audio, never a transcript, never anything you dictated. There is no field for them and the server rejects anything else.

The full detail is on the privacy page.

Troubleshooting

The installer will not run, or Windows blocks it

Which message you get tells you which of two different things is happening, and only one of them can be worked around. Both are covered under Installing: "Windows protected your PC" is SmartScreen and you can continue past it, while "An Application Control policy has blocked this file" is Smart App Control and you cannot. Neither means the download is damaged. Both are fixed by the code-signing certificate we are buying.

I hold the key and nothing happens

• Check the status pill or the tray icon to see whether Voyce thinks it is recording. If it is, the problem is the microphone; if it is not, the problem is the key.
• Another application may have claimed the same key. Set a different one.
• If your microphone is muted in Windows, Voyce notices and tells you rather than silently recording nothing.

The words appear in the wrong window

Voyce types into whatever has focus when the recording ends. If you click elsewhere mid-sentence, the text follows your click. Keep focus where you want the words.

AI formatting is greyed out or does nothing

AI formatting needs a small language model to run locally. On Windows 11, Smart App Control can block the component that loads it. If that happens the row is disabled and says so. Voyce can use a signed local runtime instead: install Ollama, then set the model and URL in the panel.

The first-run download failed

The model is fetched once and retried automatically. If it never arrives, the panel says the model is missing rather than pretending to be warming up. Check the connection and reopen the panel.

Dictation is slower than I expected

The accurate model is meaningfully slower than the fast one, and AI formatting adds time on top because it rewrites the transcript after transcribing it. If speed matters more than punctuation, switch both off.

Where Voyce keeps its files

Your settings, your licence key, the vocabulary Voyce has learned from you and the log all live together in %LOCALAPPDATA%\Voyce. The speech model you downloaded sits there too.

Back to Voyce · Choosing a dictation app · Changelog · Privacy · Terms