Windows voice input using a local Whisper model running on CPU.
wtalk is a simple Windows program with a text box and two buttons. Record your speech, stop when done, and the transcription appears in the text box. Cut the text to the clipboard to paste it wherever you need it.
- whisper.cpp cloned alongside this repo at
C:\whisper.cpp - The
base.enmodel: wtalk looks forggml-base.en.binnext towtalk.exefirst, then falls back toC:\whisper.cpp\ggml-base.en.bin. Download it from Hugging Face. - Visual Studio (for the MSVC compiler)
From a VS developer command prompt in the repo directory:
build.bat
whisper.cpp must be built first as a static library with the static CRT:
cd C:\whisper.cpp
rmdir /s /q build
cmake -B build -DWHISPER_BUILD_EXAMPLES=ON -DBUILD_SHARED_LIBS=OFF "-DCMAKE_CXX_FLAGS_RELEASE=/MT /O2 /Ob2 /DNDEBUG" "-DCMAKE_C_FLAGS_RELEASE=/MT /O2 /Ob2 /DNDEBUG"
cmake --build build --config Release
| Action | Button | Shortcut |
|---|---|---|
| Raise window and start/stop recording | — | Numpad + (global) |
| Start / stop recording | Record / Stop | F10 |
| Copy text to clipboard and clear, then minimize | Cut to Clipboard | F12 |
Transcription runs on a background thread so the UI stays responsive while the model processes audio.
Each transcription is appended to wtalk.log in the same directory as the executable, with a timestamp on each entry.