7 Commits

Author SHA1 Message Date
Your Name
baa5e2e69e Feat: Integrated Local LLM (Llama 3.2 1B) for Intelligent Correction -- New Core: Added LLMEngine utilizing llama-cpp-python for local private text post-processing. -- Forensic Protocol: Engineered strict system prompts to prevent LLM refusals, censorship, or assistant chatter. -- Three Modes: Grammar, Standard, Rewrite. -- Start/Stop Logic: Consolidated conflicting recording methods. -- Hotkeys: Added dedicated F9 (Correct) vs F8 (Transcribe). -- UI: Updated Settings. -- Build: Updated portable_build.py. -- Docs: Updated README. 2026-01-31 01:02:24 +02:00
Your Name
3137770742 Release v1.0.4: The Compatibility Update
- Added robust CPU Fallback for AMD/Non-CUDA GPUs.
- Implemented Lazy Load for AI Engine to prevent startup crashes.
- Added explicit DLL injection for Cublas/Cudnn on Windows.
- Added Corrupt Model Auto-Repair logic.
- Includes pre-compiled v1.0.4 executable.
2026-01-25 20:28:01 +02:00
Your Name
84f10092e9 Release v1.0.2: Implemented Style Prompting & Removed Grammar Correction
- Removed M2M100 Grammar Correction model completely to reduce bloat/complexity.
- Implemented 'Style Prompting' in Settings -> AI Engine to handle punctuation natively via Whisper.
- Added Style Presets: Standard (Default), Casual, and Custom.
- Optimized Build: Bootstrapper no longer requires transformers/sentencepiece.
- Fixed 'torch' NameError in Low VRAM mode.
- Fixed Bootstrapper missing dependency detection.
- Updated UI to reflect removed features.
- Included compiled v1.0.2 Executable in dist/.
2026-01-25 13:42:06 +02:00
Your Name
0b2b5848e2 Fix: Translation Reliability, Click-Through, and Docs Sync
- Transcriber: Enforced 'beam_size=5' and prompt injection for robust translation.
- Transcriber: Removed conditioning on previous text to prevent language stickiness.
- Transcriber: Refactored kwargs to sanitize inputs.
- Overlay: Fixed click-through by toggling WS_EX_TRANSPARENT.
- UI: Added real download progress reporting.
- Docs: Refactored language list to table.
2026-01-24 19:05:43 +02:00
Your Name
4b84a27a67 v1.0.1 Feature Update and Polish
Full Changelog:

[New Features]
- Added Native Translation Mode:
  - Whisper model now fully supports Translating any language to English
  - Added 'task' and 'language' parameters to Transcriber core
- Dual Hotkey Support:
  - Added separate Global Hotkeys for Transcribe (default F8) and Translate (default F10)
  - Both hotkeys are fully customizable in Settings
  - Engine dynamically switches modes based on which key is pressed

[UI/UX Improvements]
- Settings Window:
  - Widened Hotkey Input fields (240px) to accommodate long combinations
  - Added Pretty-Printing for hotkey sequences (e.g. 'ctrl+f9' display as 'Ctrl + F9')
  - Replaced Country Code dropdown with Full Language Names (99+ languages)
  - Made Language Dropdown scrollable (max height 300px) to prevent screen overflow
  - Removed redundant 'Task' selector (replaced by dedicated hotkeys)
- System Tray:
  - Tooltip now displays both Transcribe and Translate hotkeys
  - Tooltip hotkeys are formatted readably

[Core & Performance]
- Bootstrapper:
  - Implemented Smart Incremental Sync
  - Now checks filesize and content hash before copying files
  - Drastically reduces startup time for subsequent runs
  - Preserves user settings.json during updates
- Backend:
  - Fixed HotkeyManager to support dynamic configuration keys
  - Fixed Language Lock: selecting a language now correctly forces the model to use it
  - Refactored bridge/main connection for language list handling
2026-01-24 18:29:10 +02:00
Your Name
f184eb0037 Fix: Invisible overlay blocking mouse clicks
Problem:
The overlay window, even when fully transparent or visually hidden (opacity 0), was still intercepting mouse events. This created a 'dead zone' on the screen where users could not click through to applications behind the overlay. This occurred because the low-level window hook was answering 'HTCAPTION' to hit tests regardless of the UI state.

Solution:
1. Modified 'WindowHook' to accept an 'enabled' state.
2. When disabled, 'WM_NCHITTEST' now returns 'HTTRANSPARENT', allowing the OS to pass the click to the window underneath.
3. Updated 'main.py' to toggle this hook state dynamically:
   - ENABLED when Recording or Processing (UI is visible/active).
   - DISABLED when Idling (UI is hidden/transparent).

Result:
The overlay is now completely non-intrusive when not in use.
2026-01-24 17:51:23 +02:00
Your Name
9ff0e8d108 Initial commit of WhisperVoice 2026-01-24 17:03:52 +02:00