Your Voice Stays on Your Device — and a Fairer Mock Exam
Speaking practice now recognizes your voice entirely on your own device, with no exceptions — nothing you say is ever sent anywhere. It also works on far more computers than before, because we now check what the recognizer actually needs instead of guessing from your graphics hardware. The mock exam has moved to its own set of hand-checked questions, so your score reflects what you really know.
What Changed
- Speaking practice is now 100% on-device, with zero exceptions: your voice never leaves your phone or computer. Until now, some devices could quietly fall back to a cloud-based recognizer built into the browser — that path is gone for good.
- And it reaches more devices, not fewer. Eligibility is now decided by what the recognizer genuinely needs — modern browser capabilities — rather than a graphics-hardware check that wrongly ruled out many perfectly capable laptops and desktops. If your device really can't run it (very little memory, or the model fails to download), the app now says so plainly and offers a retry. It will never silently switch to a less private method.
- The mock exam is fairer now. Grammar and vocabulary sections draw from a dedicated pool of hand-checked, exam-only questions, kept separate from your daily study items — so your score reflects what you actually know, not sentences you have already drilled.
- New: furigana that fades as you learn. Once you've answered a word correctly several times and your memory of it is strong, the reading help above its kanji steps aside on regular question cards — a gentle nudge to read real Japanese on your own. Tap "show reading" any time to bring it back (the app honestly counts that as a hint), every answer still reveals readings afterwards, and an "Always show readings" option in Settings keeps furigana permanent if you prefer.
- Polish from daily study: answer feedback now stays in view with the next action right below it, resumed sessions refresh questions that expired while you were away, quiet-session settings are respected consistently, listening playback problems now show an honest error instead of failing in silence, and signing in from an external provider closes cleanly and returns you to study.
- Plus small fixes found in real sessions: cleaner page titles, better spacing on phones, and a vocabulary correction so 起きる is consistently taught in its "wake up" sense.
- Under the hood, the whole platform moved to the newest long-term foundation (.NET 10). You won't see anything different today — that's the point. It is the maintenance that keeps Mastermind fast, secure and dependable for many years, so this tool can stay with you for the long haul.