mirror of https://github.com/yusufipk/dikte.git synced 2026-09-11 10:56:10 +00:00

Files

T

yusufipek 93f3db889d Say out loud that Windows on ARM gets an emulated whisper

Checked against the release listings rather than guessed: whisper.cpp publishes
Win32 and x64 for Windows and nothing else, while llama.cpp does publish
bin-win-cpu-arm64.zip. So a Snapdragon machine gets a native cleanup model and
an emulated transcriber, which is slow enough that the cloud is the better
answer there, and neither the code nor the README said so.

The test pins it, so that a whisper.cpp release which does start publishing an
arm64 build turns the choice red rather than being quietly ignored.

2026-08-16 11:12:36 +03:00

3.1 KiB

Raw Blame History

Dikte on Windows

Press Ctrl+Space, talk, press again: what you said is transcribed, cleaned up and pasted where your cursor is.

Requirements

Windows 10/11
Python 3.11+ with PyQt6 (pip install PyQt6; install.ps1 installs it when it is missing)
ffmpeg for microphone capture: winget install Gyan.FFmpeg

Installing

powershell -ExecutionPolicy Bypass -File install.ps1

This adds a Dikte entry to the Start Menu and a dikte command to the terminal. Add -Autostart to also start it at sign-in; -Uninstall removes all of it and leaves the repository and your settings alone.

To try it without installing anything:

python dikte.py

First run

The tray icon appears and the Settings window opens.
Under API and models, download a local whisper model (the whisper.cpp Windows build is fetched automatically) or enter an OpenAI, Groq or OpenRouter key.
The shortcut defaults to Ctrl+Space and is changed under Shortcuts. While Dikte runs, Windows' own hotkey service (RegisterHotKey) listens for it: nothing to install and no permission to grant.

What is different from Linux and macOS

Meeting recording (microphone + speakers) is not supported yet. Windows does not offer what the speakers are playing as a capture device, so there is nothing to record the far side from. Everything else works, including transcribing audio and video files.
The shortcut is swallowed: while Dikte holds Ctrl+Space, the focused application does not see it. This is how macOS behaves too, and unlike the Linux listener, which shares the key.
No external tools for the clipboard or the key press: both go straight through the Windows API (the clipboard, SendInput).
Settings live under %APPDATA%\Dikte, models and recordings under %LOCALAPPDATA%\Dikte.

Performance

The local install fetches whisper.cpp's OpenBLAS build, which transcribes about twice as fast as the stock one on a plain CPU. There is no GPU build to fetch for machines without an NVIDIA card, and none for Windows on ARM either: whisper.cpp publishes x64 only, so a Snapdragon machine runs it under emulation and the cloud is the faster option there.
Setting Settings → API and models → Threads near your physical core count helps noticeably; the server's own default is 4.
If speed matters more than accuracy, ggml-small and ggml-base are much faster; ggml-large-v3-turbo-q5_0 transcribes best.

Troubleshooting

Recording does not start: does dikte doctor find ffmpeg, and does dikte devices list your microphone? devices also takes a fresh listing, which is what to run after plugging one in.
Nothing is pasted: a normal-privilege process cannot type into an elevated (administrator) window; run Dikte elevated too, or paste by hand. The text lands on the clipboard either way.
The shortcut does nothing: another application already holds the combination. Dikte says so in a tray notification when it asks for the key; pick a different one under Settings → Shortcuts.

3.1 KiB Raw Blame History