Double-tap a modifier key, say what you mean, and double-tap again. The text appears at your cursor in whatever app you were already using. Recognition happens on your own computer.
Get it from the Microsoft Store
Dictation is only useful if it fits into what you were already doing. Murmur is built around staying out of the way.
Email, chat, a code comment, a form on a web page. The text goes into the window you already had open, not into a box you have to copy out of.
Recognition runs locally. After the one-time model download there is no upload, no account, and no cloud service. It works with your network off.
The capsule never takes focus from the window you are typing in. That is the whole reason the text reliably lands where you are looking.
A bare double-tap of a modifier key. Press any other key and Murmur steps aside, so copy, paste and window switching keep working as before.
A few seconds of speech is transcribed in well under a second on an ordinary processor, so dictating a message is quicker than typing it.
The interface is available in English, Simplified and Traditional Chinese, Japanese, Korean, French, German, Spanish and Italian.
One small capsule while you speak, and one settings page for everything else.
Three steps, and only the first one happens more than once.
Open the settings window and start the one-time download. It resumes where it stopped if the connection drops.
Put your cursor where you want the words. Double-tap the trigger key and a small capsule appears while you talk.
The capsule disappears and your words are inserted at the cursor, in the app you were already working in.
Practical guides to dictating on Windows, and what to know before you pick a tool.
Built-in options, where they fall short, and a setup that types into every window.
WhyWhat happens to your voice in a cloud dictation service, and when local recognition is the better choice.
TutorialFrom first launch to your first dictated paragraph, plus habits that make dictation faster than typing.
WhyFocus stealing is the hidden reason dictation tools lose your words. Here is how to avoid it.
ComparisonCloud AI rewriting on a monthly plan, or local recognition you buy once. An honest side-by-side.
What people ask before installing.
No. Speech recognition runs on your own computer. Your audio is never transmitted, never stored after the text is produced, and never used to train anything. Apart from downloading the speech model once, Murmur never connects to the internet: no voice upload, no telemetry, no analytics. You do not have to trust anyone's promise about what happens to your recordings, because they never leave your PC.
No. There is no sign-in, no profile and no server-side component of any kind. Nothing about you is collected, because there is nowhere for it to go.
The speech recognition model is about 1.1 GB, which is too large to include in the installer. Murmur downloads it once, into your user data folder, and the download resumes from where it stopped if it is interrupted. After that it is never downloaded again.
Any application that accepts typed text. Murmur inserts the text into whichever window had your cursor, so it does not need to know anything about the app you are using.
No. The trigger is a double-tap of a modifier key on its own. The moment you press any other key while holding it, Murmur treats it as an ordinary shortcut and gets out of the way. You can also adjust how fast the two taps must be.
Yes. The settings window has a slider for how far above the bottom of the screen it sits. It also follows you across multiple desktops and works over full-screen apps.
Nothing is inserted. Silence returns an empty result rather than an invented sentence, so a gesture you did not mean to make costs you nothing.
Windows 10 or later on a 64-bit processor, a microphone, and about 1.5 GB of free disk space for the speech model. No graphics card is required — recognition runs on the processor.