IGOR SZUNIEWICZ

Fast local dictation · Windows 10/11 · v2.7.3

Mówik

Release the shortcut and Mówik starts transcribing locally at once, then inserts the result into the active window without waiting for a cloud round trip.

I am the solo creator and developer. I built the Python desktop app, its microphone buffer and local faster-whisper pipeline, the bilingual settings interface, custom commands and the Windows release process.

Solo creator and developer · Python · Windows 10/11 x64 · MIT

Get the current release Read the source
Mówik Center in Polish with shortcut, microphone, language and quality settings
Mówik Center in Polish · interface captured in v2.6.0

Dictation path

Hold F8, speak, release

F8 is the default shortcut. Releasing it starts recognition; the finished text goes to the active field, the clipboard or both.

  1. 01ShortcutHold F8
  2. 02AudioShort RAM buffer
  3. 03RecogniseLocal faster-whisper
  4. 04OutputActive field / clipboard

While Mówik runs, the microphone stays open to maintain a short pre-roll. The audio remains in memory and is discarded after transcription. This catches the first syllable without saving recordings to disk.

The first model download needs internet. After that, Mówik does not wait for an upload or cloud response: recognition uses NVIDIA CUDA in float16 when supported and falls back to CPU int8 automatically.

Interface

Everyday controls stay visible; model details stay one level deeper

The main settings expose the shortcut, microphone, language, quality profile and feedback. Model, device, speech detection and buffer controls remain in the advanced section for users who need them.

English Mówik dictation settings with shortcut, output and advanced recognition controls Dictation and local model controls · v2.6.0 capture English Mówik sound settings for recording and transcription feedback Optional recording and transcription feedback · v2.6.0 capture

The interface captures above are from v2.6.0. Custom commands were added in v2.7; the current public release is v2.7.3.

Local boundary

Audio stays in RAM; transcription stays on the PC

Local means the recognition path does not send recordings or dictated text to a remote speech service.

Microphone
It remains open while the app runs so the pre-roll can catch the start of speech.
Audio
A short buffer lives in RAM and is never written to disk by Mówik.
Transcript
It is typed or copied, and technical logs omit the dictated text.
Model
The selected model is downloaded first, then faster-whisper runs it locally.
Optional correction
Local Ollama support is off by default and connects only to 127.0.0.1.

Custom commands

Custom commands use a separate shortcut

F7 is the default. Text and open actions use exact phrase matching. Terminal actions use an explicit phrase prefix, validate the remaining words and leave the command as a clipboard draft for the user to review and run.

  1. 01

    Insert text. A command can place a predefined phrase in the active window.

  2. 02

    Open a trusted target. An app, file or HTTPS address requires confirmation.

  3. 03

    Draft a terminal command. The validated text is copied to the clipboard. Mówik does not execute it or press Enter.