1.2.0
CurrentAdded
- Metal GPU acceleration. Whisper now runs on the Apple Silicon GPU with tuned threads; large models went from about two minutes to seconds.
- A quantized Accurate model at roughly 574 MB, near the top of the accuracy range at about a third of the RAM. It is now the default.
- Per-app profiles. Blurt detects the app you are dictating into and swaps model, cleanup, vocabulary, smart match and replacements for that dictation only. No match falls back to your global settings.
- Review and edit before inserting. An optional editable box after recording: fix the text, then insert with ⌘↩ or discard with esc.
- Live dictation. Text streams to the HUD as you speak.
- Hands-free mode. Tap the trigger once to start, again to stop.
- Word replacements and smart match in Dictionary, plus custom vocabulary to bias recognition toward your terms.
- History editing, click-to-copy transcripts, and export through a Save dialog.
Changed
- Plain-language model names. The picker now reads Fast, Accurate (recommended) and Max Accuracy instead of Base, Small and Turbo.
- Six-tab layout: Home, History, Dictionary, Profiles, Settings, Permissions.
- New Flowwave app icon, regenerated at every platform size.
- The 500 MB model was removed, too close to the 574 MB one to justify keeping.
Fixed
- The menu-bar icon is now a template image, so it renders as the wave glyph and recolours for light and dark menu bars instead of showing a solid white square.
- Mode toggles apply immediately. Hands-free, live and review no longer need Save & Apply.
- Toggle switches show their on and off state; the checked-state styling had only been applied to one of them.
- The History list clamps to three lines with show more and less, and short entries stay unclamped.
- The dashboard refreshes after each dictation, not only History.
- The in-app Record button honours live mode.
- The installer restores the executable bit and retries the launch, fixing Launch Services error -10810.