from Hacker News

I vibe-coded a speech-to-text tool for Wayland that works for me. You trigger it with a keybind, speak into your mic, and it transcribes using OpenAI Whisper or Google STT (local is coming), then either types it directly into your active text field or saves it to the clipboard.

It uses PipeWire for audio capture and works signal-driven, so there's no background process running. Just on-demand transcription when you need it. I've tested it on Niri and it should work on Hyprland, though I haven't tested GNOME or KDE yet.

This was a one day Rust project and probably has some bugs since I just implemented it. It's definitely rough around the edges, but it serves its purpose for quick dictation. I'm open to feedback and input from anyone who tries it out.