I built AriaType, a local-first voice keyboard for macOS (currently in Beta v0.1.0).
The motivation: I wanted a voice input solution that: 1. Respects my privacy (no cloud processing required) 2. Works everywhere (not just in browsers or specific apps) 3. Actually produces clean, usable text (not raw speech-to-text)
How it works: - Hold a hotkey (Shift+Space by default), speak, release - Whisper-based models (or sensevoice) run locally for transcription - Optional AI polish removes filler words, fixes grammar, formats text - Text gets injected into whichever app has focus
Privacy focus: - All processing happens on-device - No internet required after initial model download - No telemetry or data collection (unless explicitly opted in) - Open Source & Transparency
Current status: - Beta v0.1.0 for macOS (Apple Silicon) - Windows version in development
Would love feedback from the HN community, especially on: - Performance and accuracy trade-offs between model sizes - Text injection reliability across different applications - Additional polish templates or features you'd find useful
GitHub: https://github.com/SparklingSynapse/AriaType
Website: https://ariatype.com/en