Double-typing / repeated characters when using ydotool on Wayland (NixOS)

[zitongcharliedeng]

Description

When using Speech Note's active-window dictation mode with ydotool on Wayland (NixOS + niri compositor), transcribed text frequently contains doubled or repeated characters/words. For example, "hello world" might be typed as "hhello wworld" or words get repeated entirely.

Environment

• OS: NixOS 24.11+ (flake-based)
• Compositor: niri (Wayland, no X11/XWayland)
• Audio: PipeWire
• Speech Note: Flatpak (net.mkiol.SpeechNote) from Flathub
• ydotool: System ydotoold with socket at /run/ydotoold/socket (0666 perms for Flatpak access)
• Model: Whisper (medium)

Steps to Reproduce

1. Install Speech Note via Flatpak on a pure Wayland compositor (no XWayland)
2. Configure ydotoold with accessible socket
3. Grant Speech Note Flatpak access to ydotool socket via overrides
4. Use start-listening-active-window D-Bus action to begin dictation
5. Speak a sentence and observe the typed output

Expected Behavior

Each character/word is typed exactly once.

Actual Behavior

Characters or words are duplicated. Appears to be a key-repeat race condition — ydotool may be sending key-down events without sufficient delay before key-up, causing the compositor to interpret them as repeated keys.

Possible Root Cause

The typing delay between keystrokes in ydotool integration may be too short for some Wayland compositors. A configurable inter-keystroke delay or using ydotool's type --delay flag could help.

Workaround

Using Handy (cjpais/Handy) instead, which uses evdev-based direct input (HandyKeys) and doesn't exhibit this issue.

[mkiol]

Thanks for reporting. Would you be able to check if increasing or decreasing Keystroke delay (Settings > Advanced) improves anything?