WinSTTSettings
The frameless WinSTT settings window for model, audio, processing, hotkeys, LLM cleanup, text-to-speech, and integrations. Changes save immediately.
Every knob in WinSTT lives in one frameless settings window — pick a model, tune voice detection, rebind the hotkey, or wire up an LLM, all from a single searchable sidebar.
The window
The settings window is frameless and resizable. The left sidebar lists the configuration panels; click one to switch. The active panel's title sits in a band that doubles as the drag region, so you can move the window from the empty header space. Close it from the sidebar — nothing is lost, because there is no "Save" or "Apply" button.
Two panels — Model and General — host sub-sections that have their own docs pages: LLM and Text-to-Speech live inside the Model panel, and the recording-mode hotkey is set in the General panel.
Pick a panel
Model
Choose the main transcription model, quantization, and device — plus the live-preview model and one-pass translation to English. LLM cleanup and Text-to-Speech are sub-sections here.
Audio
Select the input microphone and tune voice-activity detection — sensitivity and silence-stop timing.
Processing
Realtime-preview behavior, beam search and batch sizes, smart endpoint detection, and context-aware cleanup.
General
App behavior, recording mode (PTT, toggle, listen, wake word), the push-to-talk hotkey, audio ducking, the overlay, and file transcription.
Hotkey
The combo-string format for the dictation hotkey and the re-paste shortcut. The hotkey itself is bound in the General panel.
LLM cleanup
Reshape dictated text with tone presets and custom modifiers — local Ollama, OpenRouter, or Apple Intelligence. Sub-section of Model.
Text-to-Speech
Read selected text aloud with the bundled Kokoro-82M voice — 54 voices across 9 languages. Sub-section of Model.
Integrations
API keys and verification for the optional cloud providers — Ollama, OpenRouter, and ElevenLabs. OpenRouter covers cloud LLM and cloud transcription; ElevenLabs is cloud transcription. Every provider is opt-in; the app stays fully functional offline.
The sidebar also has a Dictionary panel (fuzzy corrections), a Snippets panel (text expansion), a History dashboard, and an About panel — covered on the Dictionary and Transcription history pages.
How changes are saved
Settings save automatically the moment you change them — there is no "Save" or "Apply" button. WinSTT writes them to your local app data instantly and applies whatever it safely can without a restart.
Most changes apply live, including swapping the active model. A few changes have to restart the speech engine, which reconnects the WebSocket and takes a moment:
Requires a restart
Switching the model backend family, toggling cloud vs local transcription, and a small set of startup-only audio options (sample rate, microphone lifecycle) reconnect the engine. The overlay shows a brief "swapping" state and the connection dot pulses while it comes back. Each restart-bound control is flagged on its panel's page.
Offline by default
WinSTT runs fully offline. The cloud panels under Integrations are opt-in — adding a key never sends audio anywhere until you explicitly select a cloud model. Removing a cloud key silently reverts to your last local model.