STTBar: native macOS dictation with local speech recognition
MY ROLEProject lead, software architect and Swift developer
As project lead, software architect and Swift developer, I designed and developed a macOS dictation app with local speech recognition and text entry across applications.
The app records audio, transcribes it locally using Parakeet through FluidAudio or WhisperKit, and inserts the result into the active application through macOS Accessibility. Self-hosted or Whisper-compatible endpoints can be used as an alternative.
Optional LLM post-processing with configurable prompts cleans up the text through LM Studio or an OpenAI-compatible interface. Model preloading, a recording HUD, replacement rules and a German and English interface complete the workflow. Configurable processing and protected credential storage support local and self-hosted operation.
Result: Implemented audio recording, local and server-based transcription, model preloading, optional LLM text cleanup and cross-application text entry. Developed a SwiftPM app with tests and an optional Docker Compose setup for a Whisper server.