Author's note

I vibe-coded this project. The keyboard shorcut is binded through GNOME custom shortcut UI, which then is binded to my Logitech mouse's side button through Input Remapper.

This is for personal use. I'm on Ubuntu, X11 (initially, wtype didn't work, so I switched to X11).

Usage: Click mouse's side button, talk, click again, text appears.

Customizable parts:

• STT provider.
• Activation and stop sound.
• Output directory for temporary .wav file.
• Logic for knowing is there a recording going on.
• Mechanism to type text to where the cursor in is. System-specific.

Dictation CLI

Small Go CLI to transcribe a WAV dropped into the current directory and insert the text at the cursor.

What it does

• If no *.wav present: plays a short pip and notifies "Recording ready" so you can record into this folder.
• If a *.wav exists: plays a pip, uploads the newest WAV to OpenAI Whisper, copies/transmits the transcription into the active app (xdotool/clipboard), and deletes the WAV.

Requirements

• Linux (GNOME/X11 or Wayland)
• Tools: xdotool, paplay or aplay, notify-send. For Wayland: wl-copy (preferred) or xclip + xdotool as fallback.
• Environment: OPENAI_API_KEY set.

Build

cd /home/kyle/dictation
go build -o dictate

Usage

• Bind the dictate binary to a keyboard shortcut.
• Press once to prepare recording (hear pip + notification), save a WAV into the folder, then press again to transcribe and insert.

Notes

• On Wayland the program will copy to clipboard and notify you to paste if xdotool cannot simulate a paste.