Skip to content

CHANGELOG.md

Latest commit

 

History

History
158 lines (124 loc) · 5.69 KB

CHANGELOG.md

File metadata and controls

158 lines (124 loc) · 5.69 KB
Raw

Changelog

All notable changes to VoiceAI will be documented in this file.

[1.2.1] - 2025-12-25

Added - Post-Processing Improvements 🔢

  • Sequential digit conversion - "one two three four five" → 12345 (Wispr Flow-style)
  • Enhanced currency formatting - "$30 million USD", "$10 USD", "$1 billion" patterns
  • Offline LLM activation check - Settings toggle now properly triggers offline processing

Fixed

  • Download URL - Changed to working Hugging Face URL (unsloth/Qwen3-0.6B-GGUF)
  • Model filename - Updated to Qwen3-0.6B-UD-Q4_K_XL.gguf (~405MB)
  • Offline processing not activating - RecognizeActivity now checks offline_llm_enabled setting

UI Polish 🎨

  • Ready status now truly centered with glowing ◉ icon
  • Larger toggle switches (130x60 with 52px thumb)
  • Detailed accessibility instructions in app
  • Ko-fi support link embedded in Settings → About

[1.2.0] - 2025-12-25

Added - Production Release 🎉

  • Clean UI Redesign - Modern shadcn-style white interface with Inter-style fonts
    • Light gray background (#FAFAFA) with white cards
    • Card-based layout for main screen and settings
    • Green toggle switches (100x50), dark buttons, subtle borders
  • Offline LLM Model Download - Qwen3 0.6B for fully offline AI post-processing
    • Download progress indicator with percentage
    • Model stored in app's private files directory (~400MB)
  • One-Click Groq API Setup - "Get Free API Key" button opens console.groq.com/keys
  • Accessibility Status Indicator - Green/red dot on main screen shows service status
  • Universal Text Injection via IME - Uses InputConnection.commitText() like real keyboards
    • Works in ALL apps including Microsoft Word, Chrome, etc.
    • Same method FUTO Keyboard and SwiftKey use
  • Smart Clipboard Fallback - For apps that don't support any injection method
    • Shows toast: "✓ Copied! Tap text field to paste"
  • Support Developer Link - Ko-fi link embedded in Settings → About section

Fixed

  • Ready status now properly centered on main screen
  • Word Android text injection (now uses clipboard + toast notification)

Changed

  • Removed accessibility section from Settings (now on main screen)
  • Complete UI overhaul to clean white theme
  • Reorganized main screen layout with Settings, Keyboard Setup, and Accessibility cards

[1.1.1] - 2025-12-25

Added

  • LocalLLMProcessor - Fully offline ML post-processing (no internet required)
    • Enhanced rule-based fallback with grammar fixes
    • Contraction handling (dont → don't, cant → can't, etc.)
    • Question detection for proper punctuation
    • ONNX Runtime integration prepared for TinyLlama/Phi-2
  • VoiceAIPipeline.createOffline() - New factory for offline-first operation
  • ProcessorTests.java - 40+ comprehensive unit tests for all processors

Fixed

  • Android manifest version now matches CHANGELOG (1.0.1 → 1.1.0)

[1.1.0] - 2025-12-14

Added - Major Update 🚀

  • Modular Post-Processing Pipeline: Complete refactor into 8 specialized processors
    • CourseCorrector - Wispr Flow's killer feature ("no wait" → clean output)
    • CommandInterpreter - Voice commands ("delete that", "new paragraph", punctuation)
    • RepetitionCleaner - Stutter and repetition removal
    • FillerRemover - Intelligent "uh", "um", "like" removal
    • PersonalDictionaryApplicator - FUTO-style custom words
    • NumberNormalizer - With fixed ordinal/currency edge cases
    • PunctuationRestorer - Question detection
    • CasingApplicator - Proper nouns, tech brands (ChatGPT, OpenAI, etc.)
  • Haptic Feedback - Subtle vibration on dictation start/stop
  • Voice Commands - "period", "comma", "new line", "new paragraph", "delete that"
  • LLMPostProcessor - Phi-2 integration ready (future release)

Fixed

  • Ordinal over-matching ("first thing" no longer becomes "1st thing")
  • Currency prefix doubling ("$100 dollars" → "$100")
  • Punctuation cleanup issues

Changed

  • Pipeline architecture: Chain of Responsibility pattern for clean separation
  • Processor order optimized for best output quality
  • Debug mode available for pipeline step logging

[1.0.3] - 2025-12-11

Added

  • Settings UI with iOS-style toggles
  • Version badge in Settings header
  • Enhanced personal dictionary handling

[1.0.2] - 2025-12-11

Added

  • Complete Stage 2 post-processing pipeline (PostProcessor.java)
  • Number normalization: "twenty five" → "25"
  • Time formatting: "four twenty pm" → "4:20 PM"
  • Year detection: "twenty twenty four" → "2024"
  • Ordinals: "twenty first" → "21st"
  • Ranges: "one to five" → "1–5"
  • Currency: "$100 USD", "$1,000,000"
  • Percentages: "25%", "25.6%", "0.5%"
  • Disfluency removal: "uh", "um", "you know", "like"
  • Punctuation restoration
  • Sentence casing with proper nouns
  • DictationController for FUTO-style dictation management
  • 30-second dictation limit toggle
  • Silence detection auto-stop toggle
  • Timer countdown display during dictation

Fixed

  • Preference key mismatch between Settings and DictationController
  • Personal dictionary encoding issues (now uses Pattern.quote())
  • Text injection now works reliably with all apps

Changed

  • Unified preferences to use VoiceAIPrefs
  • Enhanced README for GitHub with badges and tables

[1.0.1] - 2025-12-10

Added

  • Personal dictionary feature
  • Basic percentage and currency formatting
  • Settings activity with tile-based UI

Fixed

  • Native library loading issues
  • Package name consistency (com.voiceai.app)

[1.0.0] - 2025-12-08

Added

  • Initial release
  • Offline voice transcription with Parakeet TDT 0.6B
  • Real-time waveform animation
  • Accessibility service for text injection
  • Support for HeliBoard and SwiftKey