Changelog

System Audio Integration

• Intelligent Ducking: OneSentence now automatically lowers the volume of other applications (like Spotify, Music, or Browsers) while it is speaking or recording for maximum clarity.
• Now Playing Support: Active speech journeys now appear in the macOS Now Playing widget, allowing you to see the text being read and control playback via system media keys.

Core & Performance

• Improved Voice Processing (VPIO) coordination with Echo Cancellation (AEC) for clearer recordings in noisy environments.

Core Improvements

• Parakeet Engine Upgrade: Migrated to ONNX for faster transcription speeds and improved stability across all Apple Silicon Macs.
• Background Updates: The app now updates itself automatically via Sparkle. Updates pause if an audio recording or playback is active.
• Audio Timing: Refined the timing sequence for recordings and removed unnecessary silence padding for a more responsive experience.
• Licensing Options: Overhauled the licensing system to support a 15-day free trial, alongside dual options for perpetual licenses or subscriptions.

Bug Fixes

• Fixed an issue where audio could occasionally overlap if playback was stopped and started rapidly.
• Resolved a rare issue where a recording could result in empty data.
• Improved automatic license activation reliability when using the standard macOS installer.
• Addressed minor routing issues in the checkout flow.

Adaptive Orchestration

• Hardware Management: Added intelligent switching between the Mac's GPU (for speed) and the Apple Neural Engine (for battery efficiency) during long dictations.
• Engine Transparency: The Models tab now explicitly shows which hardware engine (Metal or Neural Engine) is currently active.

Visual Harmony

• Semantic Colors: Standardized visual feedback colors across the app (Orange for Input, Purple for Processing, Blue for Delivery).
• Unified HUD: Consolidated transcription and recording states into a single, cleaner visual feedback system.

• Audio Feedback: Refined audio device integration and added an optional "Stop" sound for clear auditory feedback when a recording ends.
• Download Resumption: Improved model download logic to better handle interrupted network connections.
• Reset Options: Enhanced the factory reset flow to ensure a clean state when performing a full reset.

• Uninstall Feedback: Moved optional uninstall feedback to a web survey to maintain strict offline data separation within the app.
• Onboarding: Added a quick reference guide to the final setup screen.
• Internal Architecture: Restructured testing and model management to support long-term updates.

• Settings Layout: Redesigned the settings interface to use a standard macOS sidebar layout.
• Maintenance Options: Added tools for a "Soft Uninstall" (removing large models while keeping preferences) and full factory resets.
• Voice Previews: Added buttons to hear samples of AI voices before downloading them.
• Clipboard Protection: Improved text extraction to safely preserve existing clipboard data when reading text from other applications.

• Protocol Architecture: Completed an internal rewrite to improve performance profiling and modularity.
• Text Handling: Enhanced parsing logic for invisible characters, emojis, and formatting structures during audio synthesis.
• Model Distribution: Migrated model downloads to Cloudflare R2 infrastructure.