릴리즈 노트 | Lumina Whisper

Release Notes

Release Notes

v0.6.1

2026-04-07

Bug Fixes

  • Replaced the bundled LLM model with an actually fine-tuned version. Improves kanji conversion and technical term correction accuracy

v0.6.0

2026-04-07

New Features

  • Bundled a fine-tuned LLM model optimized for voice input. Improves kanji conversion and technical term correction accuracy in Japanese

v0.5.0

2026-04-06

Bug Fixes

  • Improved LLM post-processing stability

v0.4.3

2026-04-06

Bug Fixes

  • Further improved LLM post-processing output accuracy

v0.4.2

2026-04-06

Bug Fixes

  • Improved LLM post-processing output quality

v0.4.1

2026-04-06

Bug Fixes

  • Fixed an issue where LLM post-processing added unwanted words to short or greeting-like text

v0.4.0

2026-04-06

New Features

  • Improved speed and stability of LLM post-processing
  • Added LLM model download and management in Settings
  • Added an LLM model preparation step to onboarding
  • Added new replacement presets: "プルリク -> PR" and "マージリクエスト -> merge request"

Bug Fixes

  • Improved context-aware text correction quality
  • Improved model download stability when retrying onboarding
  • Fixed a post-processing issue where "アジャイル" could be incorrectly altered

v0.3.2

2026-04-06

Bug Fixes

  • Improved transcription accuracy

v0.3.1

2026-04-05

Bug Fixes

  • Improved transcription and replacement accuracy

v0.3.0

2026-04-05

New Features

  • Added Sparkle-based auto update support — updates can be checked from the in-app menu
  • Distribution is now unified to direct DMG only
  • Added appcast update pipeline — signed DMG metadata can now update `https://lumina-whisper.com/appcast.xml`

v0.2.0

2026-04-04

New Features

  • First-launch setup wizard — guided flow for granting Accessibility and Microphone permissions and configuring your hotkey
  • Model warmup screen — displays progress while the model initializes on first use and safely blocks recording until ready
  • Accessibility permission and language selection added to onboarding — complete permission grant and language choice within the setup wizard
  • Default hotkey unified to Fn — new installations now default to the Fn key
  • Expanded recognition language support to 13 languages — added Chinese, Korean, French, German, Spanish, Portuguese, Russian, Arabic, Hindi, and Italian alongside Japanese and English

Bug Fixes

  • Microphone permission is now requested at launch — fixes the issue where the permission dialog appeared during the first recording
  • Introduced RMS-based silence detection — prevents hallucination text from being inserted during silent segments
  • Increased paste delay for text insertion (50ms → 200ms) — fixes occasional missed insertions
  • Model settings are now saved after download — fixes transcription engine failing to initialize on first launch
  • Fixed hit testing for language selection options — added contentShape so touch targets are correctly sized

v0.1.0

2026-04-02

New Features

  • One-key voice input with Fn — text inserted directly into the active window
  • WhisperKit-powered transcription engine — fast, accurate recognition using Apple Neural Engine
  • Direct text insertion with clipboard fallback — paste manually with ⌘V if auto-insertion fails
  • Model selection (tiny / base / small / medium) and language setting (auto / ja / en)
  • Local LLM post-processing — automatic filler removal with custom and per-app instructions
  • Text replacement rules — automatically apply custom substitutions to transcribed text
  • Word dictionary — register proper nouns to improve Whisper recognition accuracy
  • Transcription history — browse and reuse past results