Legal, Privacy, and Attributions

KanjiVG supplies the stroke diagrams used in kanji writing practice.

• KanjiVG is copyright Ulrich Apel. Source Project: https://github.com/KanjiVG/kanjivg
• PeraNabi Study ships an adapted stroke-path bundle for in-product playback and exposes that bundle from this site. Per the ShareAlike clause, the adapted bundle is itself released under CC BY-SA 3.0 — the same license as upstream KanjiVG. Changes from upstream: path coordinates normalized and re-bundled for in-app playback; stroke-order information itself is unchanged.
• KanjiVG and Ulrich Apel do not endorse PeraNabi Study or this adapted bundle. The KanjiVG name and the author's name are used here solely for attribution under CC BY-SA 3.0.

Kanji Canvas checks the kanji you draw. It runs in your browser, and your pen strokes are not uploaded.

• Kanji Canvas is copyright Dominik Klein. Source Project: https://github.com/asdfjkl/kanjicanvas
• PeraNabi Study ships the upstream browser recognizer plus a trimmed reference-pattern bundle generated from the Kanji Canvas data for in-product N5 writing practice.
• Kanji Canvas and its author do not endorse PeraNabi Study. The project and author names are used here solely for attribution.

We use AivisSpeech to make pronunciation clips and some listening audio in advance. The app plays those recordings; it does not generate speech while you study. We do not redistribute the engine or voice-model weights.

• Pronunciation audio uses にせ (ノーマル); listening dialogue validation audio uses にせ (ノーマル) and まお (ノーマル). Model page IDs: 6d11c6c2-f4a4-4435-887e-23dd60f8b8dd and a59cb814-0083-4369-8542-f51a29e72af7.
• Pre-generated WAV/MP3 audio clips produced offline under ACML 1.0 terms; used as pronunciation and listening-comprehension reference audio for language learning — a use that falls outside the ACML 1.0 prohibited-use categories.
• Generated audio is not official content and is not affiliated with or endorsed by Aivis Project, AivisHub, the model creators, or the speakers/characters represented by the voice models. Model names are used here solely for attribution.

Optional speaking practice can run a speech recognizer entirely in your browser. PeraNabi Study hosts the Transformers.js and ONNX Runtime files used for speaking checks. Whisper model files download from Hugging Face and are not modified.

• Transformers.js © Hugging Face (Apache-2.0). Whisper © OpenAI (MIT), via the onnx-community/whisper-base and onnx-community/whisper-tiny ONNX conversions. ONNX Runtime © Microsoft (MIT).
• PeraNabi Study hosts the Transformers.js and ONNX Runtime files used for speaking checks. Whisper model files download from Hugging Face and are not modified.
• Hugging Face, OpenAI, and Microsoft do not endorse PeraNabi Study; their names are used here solely to credit the software.

Speaking practice checks whether the browser recognized the target reading; it does not grade pronunciation accuracy. The converter checks the reading of the words your browser heard. PeraNabi Study hosts the converter and its dictionary.

• kuroshiro and kuroshiro-analyzer-kuromoji © Hexen Qi (MIT). kuromoji © Takuya Asano (Apache-2.0), including the mecab-ipadic dictionary © Nara Institute of Science and Technology.
• The converter checks the reading of the words your browser heard. PeraNabi Study hosts the converter and its dictionary. No dictionary data is modified.
• The authors of kuroshiro and kuromoji do not endorse PeraNabi Study; their names are used here solely to credit the software.