0.3.1

What's Changed since 0.3.0

whisper.cpp upgraded to v1.8.4

• Updated the vendored whisper.cpp submodule to the v1.8.4 release
• Added optimized BLAS support (llamafile/sgemm.cpp)
• Updated Metal shaders with new ggml-common.h and ggml-metal-impl.h headers

Push-to-Talk single key support

• The PTT hotkey can now be bound to a single key without modifiers (e.g. F5)
• Toggle hotkey still requires a modifier key combination

Transcription improvements

• Enabled suppress_nst for cleaner whisper.cpp output
• Strip bracketed hallucination markers like [Music], [BLANK_AUDIO], [Música]
• Improved emoji filtering — preserves text-like symbols (©, ®, ™) while still removing hallucinated emoji
• Auto-switch between English-only (.en) and multilingual models when changing language in Settings

Stability fixes

• Stale model-load callbacks: A generation counter prevents callbacks from cancelled model loads from updating UI state
• Hotkey teardown ordering: Hotkeys are unregistered before shutdown teardown to prevent callbacks during cleanup
• Clipboard restore: Detects if the user copied something between consecutive paste operations and refreshes the saved clipboard accordingly
• Stale paste guard: Paste dispatch checks its generation to avoid double-paste from rapid insertText calls

Housekeeping

• Added MIT LICENSE
• Updated README
• Added .gitignore, cleaned build artifacts from history

0.3.0

• Use all GPU and ANE (Apple Neural Engine) power to increase performance by more than 10x
• Language selection for multilang models
• Bug fixes

0.2.0

• You can now Copy & Paste the latest transcription
• Enable launch at login (disabled by default)
• Bug fixes

Initial release

Initial release for Voicecom