v1.5.12

# FluidVoice v1.5.12

## What's New
- Reworked dictation into two configurable shortcuts, each with its own custom prompt( AI ON ) and just raw transcription ( AI OFF) as needed. 
- Unified AI on/off and custom prompts to avoid cofusion. Now AI prompt equates to AI enhancement off. So you can toggle AI on/off on the fly with a shortcut as needed or add a separate shortcut for OFF and ON!
- Fixed overlay actions staying functional after the main settings window closes.
- Added a compact `Backup & Restore` utility in Preferences for quicker export and import access between devices or for backup!
- Added a configurable `Cancel Recording` shortcut in Settings, defaulting to `Escape`, so recording cancel behavior can be remapped.
- Added microphone switching to the menu bar for quicker input-device changes without opening Settings.



## Voice Engine Updates
- First Voice app to add `Cohere Transcribe` as a new speech model option. Very accurate with punctuations and numbers. But a little slower. Give it a try and let me know how you like it! ( 14 languages but needs manual selection)
- Added `Parakeet Flash (Beta)`, an even faster English-only local streaming model for low-latency live dictation.
- Improved Cohere performance with split Neural Engine/GPU execution and async chunk prefetch.
- Fixed Cohere model downloads and transcription failures.
- Added manual language selection for Cohere in Voice Engine settings.
- Added stronger validation for external Cohere artifacts so mismatched model contracts fail earlier and more clearly.
- Added deeper `Parakeet Flash` diagnostics for cache resolution, missing model files, and download/load verification to make stuck `Loading model...` failures easier to trace.

## File and Meeting Transcription
- Added OGG support for file transcription uploads and drag-and-drop.
- Expanded meeting transcription format support with broader macOS-native audio and video compatibility.

## Other Fixes
- Added manual backup export and import for app settings, prompt profiles, transcription history, and stats, with API keys excluded from backup files.
- Synced microphone selection state between the menu bar and Settings.
- Made `Launch at startup` more robust.
- Fixed API key authentication for localhost and other local model endpoints that still require an `Authorization` header.
- Fixed the top notch overlay so it shows the active prompt name correctly during prompt-mode recording.
- Fixed shortcut reassignment so pressing an already-assigned shortcut rebinds it cleanly instead of triggering the live recording action.
- Fixed modifier-only shortcuts so multi-modifier combos like `fn + control` can be recorded and matched correctly. 

## Credits
- Thanks to @yelloduxx for the original prompt-mode and overlay work.
- Thanks to @kabhijeet for the localhost API auth fix in PR #233.
- Thanks to @daaain for the media format support contribution.
- Thanks to all others I've missed for suggestions and contributions! 

## Need Help?
- Report issues: https://github.com/altic-dev/FluidVoice/issues