Yap is a tiny voice-to-text dictation app for Linux, Windows, and macOS. A global hotkey, a transparent floating pill, and your words land in whatever app is focused.
One hotkey, two transcription backends, and zero friction. Yap stays in the tray and shows a transparent pill when it's listening.
Hit your shortcut anywhere — terminal, browser, IDE, chat — speak, then hit it again. Yap pastes into the focused app.
Whisper-large-v3-turbo via Groq's API. Sub-second transcription for everyday dictation. Free tier is generous.
Bundled whisper.cpp via whisper-rs. Pick a model from Tiny (31 MB) up to Large Turbo (874 MB). Cached in memory after first run.
A floating, always-on-top indicator that pulses while listening. No window clutter, just vibes.
Silence is trimmed automatically; samples below 0.5% amplitude are zeroed. Cleaner audio in, cleaner transcripts out.
Tauri 2 + Rust binary. xdotool / ydotool / enigo for keyboard injection on X11, Wayland, Windows, and macOS.
Audio is captured at 16 kHz mono via cpal, gated, trimmed, and sent to your chosen backend. The transcript is written to your clipboard and pasted into whatever app has focus.
Mic recording starts at 16 kHz mono with a noise gate live on the stream.
Both ends are trimmed at a 1% threshold with 150 ms padding.
Either POST a WAV to Groq, or run whisper-rs in a blocking thread.
Transcript hits the clipboard, then a synthetic Ctrl+V fires.
Switch backends anytime in Settings. Local models live in your app data directory and load once per session.
whisper-large-v3-turbo, served fast. Bring your own key from console.groq.com.
whisper.cpp via whisper-rs. Pick a GGML model that suits your machine.
Yap is built from source with Tauri. You'll need Rust, Node 18+, and pnpm. Platform prerequisites are listed in the README.