Voice typing for Windows: dictate into any app

Windows has dictated for years. Press Win + H in any text field on Windows 10 or 11 and a small bar appears, ready to type what you say. It is free and already installed, so anything else has to earn its place.

What the built-in one gets wrong, and it is not the recogniser

Voice typing listens through whatever microphone Windows is using. On a laptop that is the array above the screen; on a desktop it is usually a webcam. Either way it is an arm's length from your mouth.

Everything follows from that distance. You speak louder than you would to a person, which is fine alone and uncomfortable in a shared room. The microphone also hears the fan and the keys. And because synthetic keystrokes pass through the active keyboard layout, dictating one language while the layout is set to another can arrive as nonsense.

Moving the microphone

Your phone is a close-talk microphone with noise suppression tuned for a mouth a few centimetres away, and it is already in your hand. It recognises the speech and hands the finished text to a small server on the PC, which types it wherever the cursor is.

Because the phone is close, a whisper carries. Because insertion goes through the Windows SendInput API as literal Unicode, the text arrives exactly as spoken regardless of layout, and your clipboard is never read or overwritten. Because the phone talks to the PC over your own Wi-Fi, nothing passes through a server of ours, of which there is none.

Where it types

Anywhere a keyboard works, since Windows cannot tell the difference. Word, Outlook, Notepad. Google Docs, Gmail and any form in a browser. Slack, Teams, Discord. Editors and terminals, for comments and commit messages. And the box where you talk to an AI assistant, which is where most people first notice they are typing less.

Prompts

Prompts are long, and that is why people write bad ones. Four sentences of context before the actual question is the difference between a useful answer and a vague one, and typing four sentences is enough friction that most of us skip it.

Speaking them costs nothing. Say the paragraph, press send. Nothing is integrated with any assistant: the text is simply typed into the field with focus, so it works the same in a browser tab and in a desktop app.

Setting it up

Install the server on the PC, the app on the phone, then pair them once with a six-digit code the PC shows you. The phone finds the computer on its own, so there is no IP address to copy. Details on how it works.

Windows 10 and 11

Both. The installer is per user, so no administrator prompt appears and no service or driver is added. There is also a single-file build if you would rather not install anything.

Where your voice goes

Recognition happens on the phone, through the Android speech service behind the microphone key on your keyboard. That service follows its own provider's policy and may send audio for processing; we have no access to it and no way to change it.

After that it is ours to answer for. The text travels to your PC encrypted with AES-256-GCM and signed in both directions, so nobody else on the Wi-Fi can read it or type into your machine. No accounts, no telemetry, no analytics in the app, and the source is public.

Questions

Is the built-in dictation any good? At a quiet desk, for short bursts, yes. Its weakness is the distance to the microphone, not the recognition.

Does Windows 11 include dictation? The same Win + H as Windows 10. This is an alternative for when the microphone distance or the keyboard layout gets in the way.

Do I need internet? The phone and PC need the same network. Recognition on the phone usually wants a connection of its own.

Does it work with Word? Yes, and Word is where the insertion method shows, because the text arrives as characters rather than as a paste. More on dictating into Word.

Is it really free? Free and open source under the GPL. A paid tier is planned for AI cleanup and translation, built on top of what already works rather than carved out of it.

macOS or Linux? The code is cross-platform and both have build scripts, but only Windows is verified end to end. Linux also needs an X11 session.