v2.5.4

Added

• New Solar (Amber → Sunset Coral) overlay theme replacing Lumina for enhanced contrast.
• Full-width Preview Mode responsive layout picker preventing vertical text squishing in General settings.
• Automatic Russian language streaming fallback in Live Preview when preferred languages match Russian.

Changed

• Decoupled floating overlay appearance completely from panel window theme to preserve high-contrast dark frosted glass.
• Multi-phrase live speech accumulator preventing recognition stalls during pauses in dictation.

Fixed

• Fixed Russian acoustic phonetic errors for goals and case agreement (e.g., proper accusative «твою цель»).
• Fixed squashed segmented buttons and labels in General settings card.
• Fixed overlay themes turning white when Light Panel appearance is selected.

v2.5.3

Added

• Live Typing / Direct Window Insertion mode with real-time cursor streaming and final Whisper replacement.
• In-app 1-click update installation and seamless restart from downloaded DMG without manual drag-and-drop.
• In-app Bug and Crash reporting card in System settings with direct Telegram delivery and diagnostic logs.
• Acoustic phoneme time-stretch expansion in Aether conditioner for studio-grade Whisper phoneme accuracy.
• Explicit Panel Theme (Dark, Light, Liquid Glass) selector in Appearance settings alongside Overlay Theme swatches.

Changed

• Moved App Theme selection to General settings while retaining Overlay Theme in Appearance settings.
• Switched live speech streaming to adaptive mode preventing stalled recognition on prolonged pauses.
• Background model pre-warming during recording for sub-150ms instant transcription.

Fixed

• Fixed Pill overlay positioning to remain centered relative to active windows and multi-monitor setups.
• Fixed Live Preview freezing on initial spoken phrases during dictation.
• Fixed permission resetting across updates by enforcing stable code signing requirements.
• Fixed duplicate lines when exporting dictated notes to Apple Notes.
• Fixed GPU stuttering and "No speech" timeouts in Orb (Magic) and ECG (Voltage) overlay modes.

v2.5.2

Changes in v2.5.2

• Dictation Modes: Added Clean, Raw, Code, Chat, and Formal dictation modes.
• Auto-Sync Open Community Vocabulary: Fully automatic 5-minute background sync with dynamic indicator.
• Local Active Document Context: Real-time Accessibility API context extraction from active editor/browser windows.

v2.5.1

What's Changed in v2.5.1

Added

• Intelligent Two-Pass Speech RMS Normalization: Dynamic level compression and soft-knee limiting for consistent speech loudness on external and distant microphones.
• Real-time Russian Grammatical Agreement & Declension: Contextual agreement rules (subject-verb agreement, preposition-case agreement, and ASR phonetic repairs).
• Anonymous Installation & Launch Telemetry: Privacy-preserving Firestore install tracking.
• Multi-Script Filtering: Clean multi-script filtering to eliminate foreign language tokens when specific languages are active.
• Dynamic Policy & Website Version Synchronization: Automatic GitHub release sync across report portal and main website.

Changed

• Menu bar popover "Start Dictation" button redesigned to cleanly match the user profile card.
• Swapped Software Updates and Audio Input sections in System Settings.
• Organic voice modulation physics for the recording preview waveform.

Fixed

• Fixed repetition loops and duplicate word runs in Live Preview.
• Optimized waveform spring animations to eliminate UI lag and high CPU usage.
• Fixed right padding and centering for the stop button on the recording pill overlay.

Full Changelog: v2.5.0...v2.5.1

v2.5.0

Added

• Top 100 User Words Frequency Engine (UserFrequencyDictionary): dynamic user lexicon tracking that learns your most-used words over time and prioritizes them in spellchecking, phonetic alignment, and candidate scoring.
• Context-Aware Russian Misrecognition Filtering: intelligent resolution for acoustic confusions (e.g. 'проявка' -> 'проверка' in testing/speech/sound contexts while preserving photography terminology).
• Strict Language Locking: automatic detection fallback that locks decoding to enabled languages when multilingual mode misdetects unselected languages.
• Audio Input Source Selector: microphone selection in Settings (System tab) and Menu Bar popover.

Changed

• Linguistic Sanitization: added automatic conversion of rogue Ukrainian glyphs and grammatical particles when transcribing in Russian mode.
• Microphone Buffer Drain: added hardware buffer flush delay before closing audio tap to capture ending syllables completely.
• Speech Boundary Padding: extended trailing audio padding from 150ms to 450ms in VAD conditioning to prevent cutting off trailing words and soft consonants.

Fixed

• Trailing Text Truncation: eliminated aggressive regex patterns and substring matching bugs in hallucination removal that previously deleted sentence endings.
• Waveform Width & Spring Physics: restored full 52-bar bouncy dynamics across the recording pill.
• Audio Hotkey Deadlock & Menu Bar Crashes: eliminated deadlock during rapid hotkey toggling and resolved MenuBarExtra presentation crashes.

v2.4.4

Added

• Audio input source selector in Settings (System tab) with automatic device discovery.
• Advanced Russian inflection and case preservation in fuzzy linguistic validation and prompt conditioning.

Changed

• Waveform audio visualization dynamic sensitivity calibrated to voice levels (-48 dB floor to -8 dB ceiling).
• Waveform bar count adjusted to 32 to eliminate edge clipping artifacts and maintain symmetric spacing.
• Moved audio input selector to the System settings tab.

Fixed

• Restored smooth 120Hz/60Hz spring physics animation across live audio waveform bars.
• Fixed hotkey deadlock during audio device initialization.
• Fixed missing environment object crash in the menu bar popover.

v2.4.3

Added

• AetherLinguisticValidator integrating native dictionary spell-checking and phonetic acoustic mishearing correction.
• Automated latest release version checking on report.html via GitHub API.
• Native feedback bridge opening device-tailored support tickets directly from macOS Settings.

Changed

• Restructured report form layout with Scribe Version at the top and inline macOS/Apple Silicon selection.
• Refined recording overlay waveform spacing and central alignment with 52 responsive bars.
• Updated security and bug policy to strictly require the latest release for issue submissions.

Fixed

• Fixed Whisper acoustic hallucinations on short conversational phrases.
• Fixed native select arrow styling on macOS web browsers with custom aligned chevrons.
• Fixed Intel and legacy platform references across Casks, documentation, and web portals.

v2.4.2

Added

• Anonymous vocabulary contribution toggle directly in Settings -> Open Community Dictionary
• Manual Sync & Upload action button with live status feedback for instant community word sharing
• 5-minute background bi-directional auto-sync between GitHub and local dictionaries

Changed

• Replaced raw word pill listings in Dynamic App Context with clean, informative domain profile descriptions

Fixed

• Restored optional community contribution controls and ensured automated local-to-remote word synchronization

v2.4.1

Added

• Multilingual subtitle and YouTube outro hallucination dictionary across 15+ languages (RU, EN, ES, DE, FR, IT, PT, ZH, JA, UK, PL, TR, KO, AR, HI)
• Language-scoped filter activation that dynamically loads hallucination blockers only for the currently active recognition languages

Changed

• Settings vocabulary and app context inspector now displays domain blocked words tailored to active user languages

Fixed

• Prevented loading unnecessary foreign language hallucination phrases outside the selected recognition scope

v2.4.0

Added

• Dynamic 7-domain application context detection (IDEs & Vibe Coding, Messengers & Chat, Notes & Writing, Browsers, Design & Creative, Crypto & Web3, General System)
• Real-time application-specific vocabulary biasing for WhisperKit and Apple Speech models
• Automatic anti-hallucination and out-of-context phrase filtering tuned per destination app
• Interactive App Context & Anti-Hallucination inspection card in Settings -> Vocabulary
• Community vocabulary terms for vibe coding, Cursor, Windsurf, Zed, and Ghostty with bilingual phonetic transliterations

Changed

• Integrated frontmost active window tracking directly into transcription conditioning and text post-processing pipelines

Fixed

• Prevented subtitle outro hallucinations and unintended boilerplate syntax across code and chat workflows