Windows' built-in dictation (Win+H) works in a lot of places, but it's cloud-based and toggle-driven. Here's a hold-to-talk alternative that works the same way in every app — and runs entirely offline.
Free & open source (MIT) · Windows 10/11 · works with any microphone
Rather than a toggle you turn on and off, VoxScribe uses hold-to-talk: hold F9 anywhere on the system, speak, and let go. The transcribed text is typed directly into whatever field currently has focus — a browser address bar, a chat window, a code editor, a document, anything that accepts normal keyboard input.
A common shortcut for "voice to text into any app" is to transcribe, copy the result to the clipboard, and let the user paste it manually. That's simple to build, but it silently overwrites whatever was already on your clipboard. VoxScribe instead simulates real keystrokes directly into the focused field, so your clipboard is left exactly as you had it.
VoxScribe auto-detects your system's default input device — including Bluetooth headsets — and records at that device's native sample rate rather than assuming a fixed one, then converts it for the speech model afterward. That avoids a common class of bug where forcing an unsupported sample rate on a given driver just crashes the recording.
Windows has a built-in voice typing shortcut (Win+H) that works in many text fields, but it depends on Microsoft's cloud speech service and isn't a hold-to-talk hotkey. VoxScribe is a separate, third-party app that runs speech recognition locally and works the same way in every app that accepts typed input.
Pasting from the clipboard would overwrite whatever you had previously copied. VoxScribe simulates real keystrokes instead, so your clipboard is never touched.
Yes, from a curated list of function keys and rarely-used keys, chosen from the tray menu. A normal typing key isn't offered, since a global hotkey on a letter key would interfere with typing that key everywhere else.