macOS dictation has one merit: it is there, all the time, in every field. We wanted the same on Linux and Windows, with better recognition and without sending our voice to a service we know nothing about. The result is called WhisperDictation, and it is free.
The gesture
Double-tap left Ctrl: French. Double-tap right Ctrl: English. You speak, text appears at every natural pause thanks to a voice detector that chunks the stream without waiting for the end of the sentence. You release, done. No window, no button.
Three engines, one setting
Not everyone has a graphics card. So we offer three modes: free cloud (Groq, with a key provided), local GPU for total confidentiality, and remote GPU — your own machine at home, queried by your laptop. At our place, the laptop dictates on the desktop’s GPU.
The traps we hit so you don’t have to
- AZERTY: simulated typing confused key codes. We go through the clipboard when needed.
- Wayland: no universal virtual keyboard. ydotool, a service and a udev rule later, it works.
- Bluetooth: a headset switching to “hands-free” mode degrades the microphone; we detect it and warn.
- Music: dictating over Spotify is unusable. Dictation pauses the music and resumes it.
- Optional neural noise suppression (RNNoise), for open spaces.
Who it is really for
People who write a lot, people who think faster than they type, and people for whom the keyboard is an obstacle. The installer guides step by step, even someone who has never opened a terminal.