目次
v1.7.0 (September 14, 2026; stable release)
The beta suffix has been removed for stable version 1.7.0. The update has been submitted for Microsoft Store certification. Availability will follow certification and publication. The current public Store version is 1.1.1; the new features below will become available there after the update is published.
- Added hold-to-talk recording while keeping the existing press-twice option.
- Recognition history now shows the actual engine and model, and removes unused rewriting status.
- Fixed first-use guide bookkeeping overwriting a newly selected model.
- Optional Soniox recognition with your own API key. Audio is sent to Soniox and separate API charges apply. Turn it off to use local Whisper.
- Frequently used terms and background context for Soniox recognition.
- Monthly and cumulative input audio duration for Whisper and Soniox, plus retrieval of overall Soniox project usage.
- Switching between microphone input and transcription of PC playback audio.
- Improvements to caret insertion, recording feedback, the usage guide, and recognition history.
v1.1.1.0 (submitted August 20, 2026; published)
This update is published on Microsoft Store. The app displays v1.1.1 and the package version is 1.1.1.0.
- Improved the reliability of recognized text output at the caret.
- Handles focus information more reliably after ordinary clicks and in some applications, reducing cases where a valid destination was incorrectly treated as unavailable.
- Speech recognition, the correction dictionary, pricing, and license terms are unchanged.
v1.0.0.0 (August 19, 2026)
This update moves Meizin Voice to its formal version 1.0 release. The app displays v1.0.0, and the Microsoft Store package version is 1.0.0.0.
- Fixed an issue where recognized text was not inserted at the caret in some editors, including Sakura Editor and Meizin Note.
- Changed the input safety guard so normal recognition output continues, pausing only immediately after an operation that creates a text selection.
- Improved live output reliability and reduced the delay before confirmed phrases appear after speaking stops.
- Corrected the release build configuration and verification process so live output is always included in product builds.
v0.3.0 (June 3, 2026)
- Translation Mode added — Speak in Japanese, get English text output automatically (with Large model selected)
- AI model lineup changed to 2 models (Small / Large-v3). Former Medium model discontinued
- Built-in dictionary expanded to 391 entries (up from 225)
- Microsoft Store submission
- Privacy Policy page published
v0.2.0 (May 30, 2026)
- Q&A Forum added
- Contact form added
- Review system added
- Help tab added (links to usage guide, Q&A, contact, updates)
- App navigation menu added (shared header for LP, Q&A, Contact, Updates)
- First-launch guide added
- Startup warmup added (eliminates first hotkey lag)
- Custom dictionary search, sort, and inline editing
- Recognition history tab improvements (text wrapping, adjustable limit, separators)
- Shift+Enter shortcut (add/save dictionary entries)
- Version display added
- Demo video recorded and embedded in LP
v0.1.0 (May 25, 2026)
Initial release.
- Local AI speech recognition (Whisper small + Vulkan GPU)
- Custom dictionary + built-in dictionary
- One-key voice input (F9, customizable)
- Japanese / English UI support
- Dark mode support
- Automatic GPU detection + CPU fallback
- Trial period (15-day free trial)
- License key activation (Lemon Squeezy integration)
- Terms of Service and Privacy Policy