Press Right Ctrl, speak, release. Lven Instant types what you said, with the punctuation, into whatever window has focus — Word, Outlook, Chrome, Slack, your editor. The speech model runs on your CPU, so your voice never leaves the machine.
SHA-2561beff6912cb77c3c0d0610a31fdf9960873012d734cbce47c5ed127d18c6ed3c.sha256 file
Lven-Instant-windows-0.8.41-Setup.exeLven-Instant-windows-0.8.41-Setup.exeIt is fast enough that the text is there when you stop talking. On consumer-grade hardware the shipped engine runs about six times faster than you speak, with a median of 0.93 seconds per clip and a 3.76% word error rate on clean speech — measured, dated and published under Research.
On the same hardware whisper-large-v3 ran 10 to 30 times slower per clip and never reached realtime.
It works in every application with a text field. The floating bubble sits at the edge of the screen; the global hotkey works from any window. Where an application refuses automated input, the transcript waits on the clipboard so nothing is lost.
It stays light. Nothing is metered because nothing goes to a server, about 600 MB of memory is all the engine needs, and no GPU is involved.
Dictate in 25 European languages, detected automatically.
Windows checks a new installer once. Here is exactly what happens.
SmartScreen scores a publisher by how many people have run its files. A new publisher is unknown, not unsafe — the warning is about reputation, not about what the installer contains.
64-bit. Tested on Windows 10 and Windows 11.
Any microphone recognized by Windows — laptop mic, USB, or Bluetooth headset.
The speech model downloads once at setup. Signing in and access checks use the network; your voice never does.