Voice Answers for WaniKani: Unofficial Tool

An unofficial tool: answer WaniKani reviews by voice. Speech recognition runs entirely on your device.

Say your WaniKani answers instead of typing them. Hold Shift, say the meaning or the reading, and release: the answer is filled into WaniKani's answer box, ready for you to submit. PRIVATE BY DESIGN • Speech recognition runs inside your browser, on your own computer. Your voice is never sent anywhere: the extension makes no network requests at all, and its speech models ship inside the extension. • It never reads the question. It only checks whether WaniKani is asking for a meaning or a reading, so a wrong answer goes in exactly as you said it. • No account, no tracking, no data collection. FEATURES • English meanings and radical names (Moonshine, fast, by default; or OpenAI Whisper, accurate). • Readings in hiragana, from a speech model that writes kana directly, so it never has to guess a reading from kanji. • Alternatives: if it also heard something else (hand / and, じん / にん), press 1–3 or click to switch before submitting. • Push-to-talk (Shift, configurable) or hands-free; the panel's mic button also works (hold it to talk). • Fill only, or fill and submit; optional auto-advance (or only after a correct answer). • Works in reviews and lesson quizzes. • Advanced: load your own fine-tuned model per language. Tools to fine-tune on your own voice are in the source repository. REQUIREMENTS No GPU needed: everything runs on the CPU. On a fast desktop answers take about 0.1 s (English, fast), 1.5 s (English, accurate) and 0.4 s (readings); slower computers take longer. The models use about 430 MB of memory while a review is open (720 MB with accurate English). Chrome 116 or newer; the download is about 140 MB. Avoid system noise gates or suppressors on the microphone: they cut off the start of words. NOTES An unofficial tool: not affiliated with or endorsed by WaniKani or Tofugu. Open source (MIT): https://github.com/jsmrcina/wanikani-voice. This extension was created with AI: all of its code was written by Anthropic's Claude Opus 5.5 (in Claude Code), directed and tested by a human.