AnkiMining 0.1.2

Small update: a setting for the window stacking and a better card layout on wide screens.

What changed

• Keep the window above the other applications is now a setting. In Settings → Application. It is on by default, as before. Turn it off and the system decides the stacking: an application opened later appears above AnkiMining instead of below it. The popup and the region selector still show over the text they belong to.
• The explanation of the card no longer stretches on ultrawide screens. In the note type the app creates, the meaning is a left-aligned block, centered on the card, at most 700 px wide. A note type you already have keeps its styling; to get the same layout, open it in Anki (Tools > Manage Note Types > Cards > Styling) and replace .meaning { margin: 12px 0; } with .meaning { max-width: 700px; margin: 12px auto; text-align: left; }.
• README rewritten around what the app does.

Install and update

Installed apps offer this version by themselves (a banner with Install and restart). Or download it: Windows AnkiMining_0.1.2_x64-setup.exe (not signed by Microsoft: SmartScreen may warn, choose More info > Run anyway), Linux AnkiMining_0.1.2_amd64.AppImage.

Credits

Voice: VOICEVOX:春日部つむぎ. Dictionary: JMdict (EDRDG, CC BY-SA 4.0). Pitch accent: Kanjium (CC BY-SA 4.0). OCR: MeikiOCR models (LGPL-3.0). Frequency counted from Tatoeba sentences (CC BY 2.0 FR).

AnkiMining 0.1.1

Small update: the audio of the cards moves to the back.

What changed

• Audio on the back of the card. In the note type the app creates, the word and the sentence read aloud now play on the back, right below the reading (hiragana) and above the meaning. The front no longer plays any audio, so it does not give the answer away. The cards you already have keep their templates; to get the same layout in a note type you already use, remove {{SentenceAudio}} from its front and add it to its back after {{Reading}} (README, "Spoken word and sentence"). Changing templates makes Anki ask once for a full sync with AnkiWeb, so the app never does it for you.

Install and update

Installed apps offer this version by themselves (a banner with Install and restart). Or download it: Windows AnkiMining_0.1.1_x64-setup.exe (not signed by Microsoft: SmartScreen may warn, choose More info > Run anyway), Linux AnkiMining_0.1.1_amd64.AppImage.

Credits

Voice: VOICEVOX:春日部つむぎ. Dictionary: JMdict (EDRDG, CC BY-SA 4.0). Pitch accent: Kanjium (CC BY-SA 4.0). OCR: MeikiOCR models (LGPL-3.0). Frequency counted from Tatoeba sentences (CC BY 2.0 FR). Licenses in the README.

AnkiMining 0.1.0

First public version of AnkiMining: hover over Japanese text anywhere on the screen to see the word, then add it to Anki.

What it does

• Lookup anywhere: hold Shift (or rest the cursor) over any Japanese text on screen (video subtitles, web pages, games, players) and a popup shows the word with reading, meanings, pitch accent and frequency. No browser extension is needed. Pause the video first.
• Words you already have: every word of the sentence is underlined green when its card already exists in Anki.
• One click to Anki: each entry has Add to Anki (or press Alt+M). The card gets the sentence, a screenshot, the word and the sentence read aloud by an offline neural voice, pitch accent and frequency. Decks can depend on the program the word was found in (profiles).
• Listen: every entry has a ▶ button that reads the word aloud.
• mpv: exact subtitles and exact audio of the line when mpv's IPC is on.
• Updates: the app looks for new versions on GitHub and offers to install them (signed files only).

Install

• Windows 10/11: run AnkiMining_0.1.0_x64-setup.exe. It is not signed with a Microsoft certificate, so SmartScreen may warn: choose More info > Run anyway.
• Linux (X11): chmod +x AnkiMining_0.1.0_amd64.AppImage && ./AnkiMining_0.1.0_amd64.AppImage.
• Anki with the AnkiConnect add-on must be open to add cards.

Checked

Both versions were tested by running the installed app's self-test (dictionary, tokenizer, OCR, voice, screen capture, pointer) and the unit tests: the Windows build on a real Windows machine (GitHub Actions), the Linux build in Docker. Adding cards has been tried against a mock AnkiConnect, not a real Anki.

Known limitations

Linux needs X11 (not Wayland). Overlapping subtitle layers and rare kanji can be misread by the OCR. See the README.

Credits

Voice: VOICEVOX:春日部つむぎ. Dictionary: JMdict (EDRDG, CC BY-SA 4.0). Pitch accent: Kanjium (CC BY-SA 4.0). OCR: MeikiOCR models (LGPL-3.0). Frequency counted from Tatoeba sentences (CC BY 2.0 FR). Full list and licenses in the README.