3.3.4

Changelog

3.3.4 (2026-04-14)

Stability & Initialization:

• App Restart Race Condition: Resolved a critical race condition that could occur during an application restart (e.g., after resetting to factory defaults). This ensures backend and frontend states remain perfectly synchronized during reboot.
• Service List Standardization: Enhanced the initialization logic for service provider lists (Translation, TTS, STT, and AI). The app now pre-seeds canonical defaults into storage during startup, preventing individual settings pages from accidentally overwriting the list with empty values if visited before the main Service page.
• Config Sync Reliability: Optimized the configuration file watcher to prevent application hangs on Windows, particularly during first-launch or after a full data reset.
• UI Refinements: Minor label and layout fixes in the Local Models and Voice Input configuration pages.

3.3.3 (2026-04-14)

Live Interpreter (formerly Monitor):

• Branding Update: The "Monitor" window has been officially renamed to "Live Interpreter" ("Phiên Dịch Trực Tiếp") to better reflect its core functionality for meetings and live audio.
• Narration Panel Redesign: Complete overhaul of the Narration (Push-To-Talk) panel with a modern layout consistent with the Context Panel. Reordered sections for a more logical workflow: Device → PTT → Options → Polish → Zoom.

PTT (Push-To-Talk) Enhancements:

• PTT Independence: Fixed a critical conflict where global Voice Anywhere hotkeys were triggering Monitor PTT events. The two systems are now strictly isolated on independent event lanes.
• PTT Configuration Expansion: Added a dedicated sub-panel for PTT settings within the Live Interpreter:
  • Polishing: AI-powered refinement for dictated speech.
  • TTS Speed: Adjustable playback speed for narrated audio.
  • Custom Prompt: Support for custom AI instructions for the polishing step.
• UI Stability: Resolved UI state bugs in the Live Interpreter toolbar and labels where PTT/Recording status could become desynchronized during rapid interaction.

Tray & i18n Refinements:

• Audio Monitoring UI: Updated Vietnamese translations and labels for audio monitoring items in the system tray menu.
• Unified Naming: Standardized the use of "Phiên Dịch Trực Tiếp" across the entire application interface.

3.3.2 (2026-04-13)

ONNX STT Engine — Windows Compatibility (Comprehensive):

• Module import fix: server.py now inserts its own directory into sys.path at startup, bypassing the PYTHONPATH-ignored behaviour of Python embeddable packages on Windows. Resolves ModuleNotFoundError: No module named 'capabilities' (and any other sibling-module import failures) that crashed the server immediately on launch.
• UTF-8 stdout/stderr: Embedded Python on Windows defaults to CP1252 encoding; server.py now forces UTF-8 on stdout/stderr so JSON output containing Unicode characters (model names, transcripts) is transmitted to Tauri without UnicodeEncodeError.
• asyncio event loop policy: Python ≤ 3.11 on Windows defaults to ProactorEventLoop; server.py now sets WindowsSelectorEventLoopPolicy before uvicorn starts, preventing startup failures on certain Windows configurations.
• Hidden console windows (install phase): Added CREATE_NO_WINDOW flag to the get-pip.py and pip install subprocess spawns in the ONNX engine installer — no terminal flashes during engine installation.
• Hidden console window (sidecar start): Added CREATE_NO_WINDOW to the Local Sidecar process spawn in local_sidecar.rs, matching the existing flag in onnx_engine.rs.
• Cross-platform setup.py: Removed import select (Unix-only module, crashes on Windows). Fixed get_default_env_dir() to use %APPDATA% on Windows. Added _venv_python() / _venv_pip() helpers returning Scripts\python.exe / Scripts\pip.exe on Windows (vs bin/python3 / bin/pip3 on Unix). Extended find_system_python() with Windows %LOCALAPPDATA%\Programs\Python\* paths and python / python3 from PATH.

3.3.1 (2026-04-13)

Out-of-the-Box Experience — Zero Config Required:

• Default hotkeys registered on first launch: All six global shortcuts are now active immediately after installation — no manual setup needed. Defaults: Ctrl+Alt+Q (Selection Translate), Ctrl+Alt+W (Input Translate), Ctrl+Alt+R (OCR Recognize), Ctrl+Alt+O (OCR Translate), Ctrl+Alt+M (Realtime Translate), Ctrl+Alt+V (Voice Anywhere). Fully reassignable or clearable in Settings → Hotkey.
• Service providers pre-seeded on fresh install: Translate, TTS, Transcription, and AI service lists now populate with sensible defaults on first run rather than showing an empty list. Translate: Transkit Cloud + Google + Bing. TTS: Transkit Cloud + Google TTS + Edge TTS. Transcription: Transkit Cloud STT + Deepgram. AI: Transkit Cloud AI. Free providers (Google, Bing, Edge TTS) work immediately without any API key.
• Consistent defaults across all windows: Fixed a race condition where the Translate window, Voice Anywhere, and Config pages each used different default values for the same service list keys — whichever rendered first would win and under-populate the list. All entry points now share the same defaults.

Reset to Defaults — Settings → About:

• New "Reset" button in the System section of the About page.
• Two-option confirmation modal:
  • Reset — Keep API Keys: Wipes all app settings and resets to factory defaults while preserving every service credential (API keys, auth tokens, custom service configs). Service instance configs stored as key@id pairs or plain objects are retained automatically.
  • Reset Everything: Clears the entire config store, returning the app to a completely fresh state.
• App restarts automatically after reset; default hotkeys and service lists are re-applied on first launch.

3.3.0 (2026-04-13)

ONNX STT Engine — Windows Support:

• Windows PATH fix: The ONNX engine process now inherits the correct Windows system PATH instead of overriding it with macOS/Linux paths, allowing Python and native sherpa-onnx DLLs to load correctly.
• Hidden console window: Added CREATE_NO_WINDOW flag on Windows so no terminal flashes when the ONNX server starts.
• Windows bundle path: Added the correct resource path (scripts/local_sidecar/server.py next to the .exe) for production Windows builds.
• Crash diagnostics: Python stderr is now captured and forwarded to the UI via the onnx-engine://stopped event so users see the actual error message when the server fails to start.
• UI error banner: Start failures (invoke error or process crash) now display a red error banner below the Start/Stop buttons instead of silently logging to console.

3.2.9 (2026-04-13)

Voice Anywhere — Tray & Config Improvements:

• Favorite Target Languages: New "Favorite Targets (Tray)" section in Voice Input settings. Configure a personal list of target languages; they appear instantly in the system tray for one-click switching without opening settings.
• Tray: Translate to (replaces Language): The system tray Voice Anywhere submenu now shows a "Translate to" section (No translate + favorites) instead of the rarely-changed source language selector — putting the most-used action one click away.
• Tray Event Sync: Changing the target language from the tray now emits voice_anywhere_target_language_changed so the Config window reflects the selection immediately without a restart.

Tray Reliability & i18n:

• Full Tray i18n for Voice Anywhere: All 12 tray submenu labels (STT Provider, Inherit from Monitor, Translate to, No translate, After Stop, Copy/Paste, Inject Mode, Replace, Append, Configure…) are now localised in all 12 supported tray languages (VI, ZH-CN, ZH-TW, JA, KO, FR, DE, RU, PT-BR, FA, UK, EN).
• Error-safe tray handlers: Replaced all .unwrap() panics in tray event handlers and update_tray() with warn!-logged error paths — set_menu, set_tooltip, all emit_all calls, unregister_all, and open log dir no longer crash the app on failure.

i18n — 22 Locales:

• Added common.add, config.voice_input.fav_targets, and config.voice_input.fav_targets_desc to all 22 locale files with native translations.

Documentation — Keyboard Shortcuts:

• Added a Keyboard Shortcuts section to all 4 READMEs (EN / VI / CN / KR) documenting the branded Ctrl+Alt+ hotkey set and explaining cross-platform compatibility (Ctrl+Alt = Ctrl+Option on macOS, works with one registration across all OS).
• Recommended defaults: Ctrl+Alt+V (Voice Anywhere), Ctrl+Alt+M (Realtime Translate), Ctrl+Alt+Q (Selection Translate), Ctrl+Alt+W (Input Translate), Ctrl+Alt+R (OCR Recognize), Ctrl+Alt+O (OCR Translate).

3.2.7 (2026-04-11)

ONNX STT Engine & Models:

• Enhanced Model Detection: Fixed logic to correctly identify and load various Zipformer model structures (standard ONNX vs shared sidecar weights).
• Improved Downloader: Refined HuggingFace model downloading with better error handling and progress reporting.
• Engine Reliability: Optimized the local sidecar engine for more stable WebSocket connections and faster initialization.

UI & Refinement:

• Config UI Modernization: Refactored the ONNX STT configuration page with a premium vertical layout using NextUI components, consistent with the app's design system.
• Standardized Inputs: Improved usability of STT provider settings with better label placement and aligned action buttons.

3.2.6 (2026-04-10)

Voice Anywhere & Dictation:

• Native Audio Feedback: Restored start/stop recording sounds by bypassing browser restrictions using native Rust audio playback. Includes a new toggle in settings to enable/disable sound effects and increased volume for better noticeability.
• Integrated Translation: Added Target Language support to Voice Anywhere. Speak in one language and have the app instantly translate and inject/pasted the result into any application.
• AI-Powered Polishing: New "Polish" mode that uses AI to refine your dictated transcripts for grammar, tone, and professional clarity before they are injected.
• Cross-Platform Pasting: Improved the reliability of automatic text pasting across macOS and Windows with intelligent focus-target capturing.
• Dictation Service: Implemented a new dictation engine using Edge Function WebSocket proxies for high-speed, short-form speech recognition.

Local Sidecar & Offline Support:

• Offline AI Logic: Full integration with the Transkit Local Sidecar, enabling private, offline LLM (Llama 3), STT (Whisper), and TTS services.
• Granular Control: New settings to filter which local components are active and a quick-access button to manage the model cache directory.
• System Prompts: Local AI models now support custom system prompts for tailored assistant behavior.

UI & Core Enhancements:

• STT Lifecycle Standard: Standardized the finalize() method across all transcription providers to ensure clean session termination and consistent results.
• Protocol Normalization: Refined how the frontend handles transcript events from various providers, ensuring a unified experience regardless of the selected service.
• Config UI Refinement: Improved the layout and responsiveness of the configuration windows.

3.2.5 (2026-04-06)

Voice Anywhere & Global Inclusion:

• Voice Anywhere FAB: Introduced a global floating action button (FAB) for instant voice-to-text input. Dictate anywhere and inject text directly into any active application (Zoom, Teams, Slack, etc.).
• Realtime Caption Overlay: Added a dedicated caption window that displays live transcription results right above the FAB as you speak.
• Smart Visual States: The FAB features 5 dynamic states: Idle (swirling pulse), Listening (live amplitude bars), Processing (spinner), Injecting (success flash), and Error (red alert).
• Customizable Aesthetics: New configuration option for Idle Button Color, allowing the FAB to blend perfectly with your desktop theme.
• Drag & Toggle: Intelligent interaction design that distinguishes between a quick tap (toggle recording) and a long-press/drag (reposition window).

Realtime Monitor & PTT Enhancements:

• Push-to-Talk (PTT) Toggle: Dedicated toolbar button in the Monitor window to quickly enable/disable PTT mode.
• Live Service Switching: Support for switching transcription services (e.g., Deepgram to Gladia) on-the-fly without stopping the session.
• Modern Log Aesthetics: Updated the Monitor log bubbles with consistent, blue-tinted visual styles for improved clarity and readability.
• Initialization Guards: Added start-up concurrency guards to prevent multiple simultaneous session requests, ensuring stability during rapid toggling.

Account & Subscription Management:

• STT Addons & Team Plans: Enhanced the Account dashboard to support purchasing STT minute bundles and managing Team-based collaborative plans.
• Subscription Awareness: Added dedicated UI for tracking plan expiration dates and quotas for STT, TTS, and AI credits.
• Secure Sign-out Flow: New SignOutButton with a confirmation modal to prevent accidental logouts. Sign-out now correctly targets local session scope for better multi-account handling.

Core & Configuration:

• Hotkeys & Voice Input Config: New dedicated configuration pages for global Hotkeys and specific Voice Input settings, including sensitivity and delay controls.
• Source Area Refinement: Improved focus handling and text injection logic to ensure cursor stability when pasting dictated text.

3.2.4 (2026-04-01)

UI & UX Enhancements:

• Header Theme Toggle: Added a compact sun/moon switch directly in the Config window header for instant light/dark mode switching.
• Config Window Refactor: Increased sidebar width to 250px for better readability and improved title/control alignment.
• Export Transcript: New "Export transcript" action in the Monitor window to save your session snapshots.
• Improved i18n: Added "Session expired" handling and refined labels for account usage.

TTS & Narration Expansion:

• Google Cloud TTS (Chirp3-HD): Integrated professional-grade TTS support with full control over Voice, Speaking Rate, and Pitch.
• Enhanced Windows Narration:
  • Full support for VB-Cable on Windows to route audio to Zoom, Teams, or Google Meet.
  • Added step-by-step setup hints and direct download links for VB-Cable in the settings.
  • Dynamic checklist labels that reflect the selected virtual device ({{device}}).

Documentation & Credits:

• README Restructure: Simplified the top-level description and moved project credits (fork from Pot Desktop) to a dedicated "Credits" section, emphasizing the significant improvements and modifications made.

3.2.3 (2026-03-31)

Narration & Audio Routing:

• Narration Mode (PTT) — Now supporting macOS & Windows: Integrated Push-To-Talk support for virtual microphone output—perfect for translating and narrating directly into Zoom, Google Meet, Teams, etc.
• Virtual Audio Setup: Step-by-step guide and test tone generator for routing audio via virtual drivers (e.g., BlackHole on macOS, VB-Audio on Windows).
• Virtual Mic Checklist: New guides for routing conference mic to Transkit's virtual output.

AI Assistant & Context Insights:

• Expanded AI Assistant Panel: Added a dedicated sidebar for real-time analysis with Smart Context Insights:
  • Modes: Suggest Answer, Quick Insight, Summarize, and Follow-up.
  • Analysis: Automatic extraction of Key Points, Next Steps, and Questions to Ask from conversations.
• Improved AI Regeneration: Support for regenerating suggestions on the fly.
• Language Flexible AI: AI can now reply in both languages (Source/Target) for better cross-cultural understanding.

Source Area & Dev Productivity:

• Redesigned Source Area: Refactored the core translation input component for improved performance, synchronous state recovery, and smarter focus handling.
• Variable Name Transformer: New shortcut (Alt+Shift+U) to instantly toggle selected text between snake_case, kebab-case, camelCase, PascalCase, and more.
• Custom STT Provider: Support for arbitrary WebSocket-based transcription services with custom connection parameters (WS URL, API Key, Delay).

Visual & Stability Polish:

• GIF Demonstrations: Updated READMEs with high-quality animated guides for Clipboard Listening, Screenshot OCR, and Translation.
• Improved Auto-save UX: Added "Open saved file" and "Open transcripts folder" actions trực tiếp từ Monitor toolbar.
• Cloud Infrastructure: Deployed stable get-cloud-credentials edge functions for seamless Transkit Cloud session management.

3.2.2 (2026-03-28)

Audio Monitor & STT Enhancements:

• Custom Vocabulary Support: Added support for custom vocabulary in Deepgram and Gladia to improve recognition of specialized terms.
• Advanced Controls: Exposed endpointing and threshold controls for Transkit Cloud transcription.
• Improved Auto-save UX: New labels and tooltips for auto-save status; added quick-open actions for saved transcript files and folders.

UI & Core Fixes:

• Source Area Redesign: Refactored the translation source area with improved focus handling, state recovery, and variable name transformation.
• Font & Style Adjustments: Standardized source text font size to 14px and optimized configuration row spacing.
• Stability Fixes: Normalized boolean configuration values and improved window event synchronization.

Documentation:

• Visual Guides: Added the "Realtime Audio Monitor" featured row and updated example GIFs in all README files.

3.2.1 (2026-03-27)

Transkit Cloud & Account:

• Major i18n Expansion: Complete English and Vietnamese translations for Transkit Cloud account management, including plan levels (Free Trial, Starter, Pro) and detailed usage tracking for STT, TTS, and AI.
• Built-in Cloud Providers: Introduced Transkit Cloud as a zero-config provider for Translation, TTS, AI, and STT (powered by Gemini 2.5 Flash and ElevenLabs).
• Quota Management: Real-time tracking of transcription minutes, TTS characters, and AI requests with automatic "Quota Reached" notifications and upgrade paths.
• Authentication Conflict Fixes: Added automatic retry prompts and informative states for concurrent auth token requests.

User Interface & Experience:

• Sorting Options: Transcripts can now be sorted "Newest at top" or "Newest at bottom" with persistent UI labels.
• Improved Error Messaging: Dynamic "Missing API Key" errors that guide users directly to the relevant provider settings.
• Auto-save Status: Clearer "Auto-save ON" indicators in the Monitor toolbar.
• Subtitle Mode (SubMode): New visual documentation and guide for the minimized overlay mode.

Distribution & Installation:

• Official Homebrew Support: Install via the new tap: brew install --cask transkit.
• Winget Integration: Verified and updated instructions for installing via winget install TransKit.
• Provider Transparency: Added clear credit information for Soniox, Deepgram ($200 credit), and Gladia (10h/month free).

3.2.0 (2026-03-25)

Documentation & Global Support:

• Default Multi-language User Guides: Standardized English (docs/user_guide.md) as primary and Vietnamese (docs/user_guide_vi.md) as supplementary guides.
• Enhanced macOS Support: Added detailed instructions for unsigned app workarounds (xattr -cr) and a step-by-step guide for Screen/Audio Recording permissions.
• Developer Options: Added instructions for building in Local-only mode (VITE_DISABLE_CLOUD=true) to bypass Supabase/Cloud dependency.
• Visual Aids: Integrated high-quality UI screenshots into both READMEs and User Guides.

Major Improvements (Refining 3.1.x):

• New Realtime Transcription Providers:
  • Deepgram: Fast STT with $200 free credit; supports built-in Google Translation.
  • Gladia: WebSocket live streaming with 10 free hours per month; features speaker diarization.
  • Soniox: Low-latency transcription for professional meetings.
• Realtime Monitor Enhancements:
  • AI Suggestion & Context: Tightly integrated LLMs (Gemini, OpenAI, Claude) to solve specialized terminology in real-time.
  • SubMode: Overlay subtitle mode for movies and video calls.
  • Smart Sorting: Toolbar options to sort transcripts newest-first or oldest-first.
  • Placeholder Redesign: Quick-access provider switching and direct session start.
• Core & Performance:
  • Installer Optimization: Reduced installer size by 20MB by removing heavy local detection engines.
  • Cloud Session Tracking: Live countdown for trial usage and account-based synchronization.
• TTS Upgrades: Support for Edge TTS (free/built-in), ElevenLabs, and OpenAI-compatible streaming TTS.
• Clipboard Sync: "Monitor Clipboard" feature to automatically translate any copied text in the background.

---

For more details, visit the Full User Guide.