STTBar

STTBar turns your voice into finished text, in any app on your Mac.

Press a hotkey and speak. Press it again and STTBar transcribes the recording, on your Mac or on a server of your choice. The text appears where your cursor is: in an email, a chat, your editor or the prompt field of an AI assistant.

What STTBar does: • Three hotkeys: cleaned-up text, raw transcript and English output (speak German, paste English). • On-device speech recognition with Parakeet or WhisperKit. After the model download, recognition works offline. Local models are preloaded when the app starts. • Or a Whisper server: on a computer in your network or on the internet, with an API key if it needs one. • Optional cleanup with a language model of your choice: locally, for example in LM Studio or Ollama, or hosted through OpenRouter and other OpenAI-compatible services. It removes filler words and gives the text a clear structure. Cleanup mode can fall back to the raw transcript if enabled. English output requires a working language model endpoint. • API keys stay in your Mac's keychain and only go to the server they belong to. • A small overlay with a live waveform shows whether STTBar is recording, transcribing or cleaning up. • Word replacements for names and jargon, so they are always spelled right. • History with privacy controls, and a sensitive mode that never stores a dictation. • German or English interface.

No account, no subscription, no ads. Recordings and texts leave your Mac only if you add a server or service yourself.

STTBar needs access to the microphone and to Accessibility. Accessibility is required to insert the text into the active app.

Requires macOS 14 and a Mac with Apple silicon. On-device recognition downloads model files once. The first download and preparation can take 1 to 5 minutes.

Sygnały wykrywalności