nihongo-it-anki

Multi-deck Anki generator for IT Japanese. AI-generated audio, pitch accent coloring, and 3 card types designed for real workplace skill building. Open source.

Decks

IT Vocabulary (it-vocab)

1516 sentences across 15 tiers. Casual to keigo workplace Japanese for software engineers.

Tier	Count	Focus
1	150	Daily essentials, git, basic actions
2	200	Agile, APIs, databases, testing
3	250	Code review, architecture, AWS
4	200	Security, debugging, documentation
5	100	Communication, soft skills
6	115	Presentations (including formal keigo)
7	30	Job interview (full keigo register)
8	30	Problem-solving discussions
9	50	AI & Agents
10	51	Documentation & Engineering Culture
11	100	Casual chat (plain-form Slack/huddle)
12	100	Live meetings (interrupting, clarifying, estimating)
13	60	Corporate life (attendance, expenses, reviews, 1on1)
14	40	Business email (keigo formulas)
15	40	Written tickets, commits, PRs (formal written register)

IT Kundoku (it-kundoku)

89 sentences across 3 tiers. Compressed single-kanji Japanese for token-efficient AI conversation.

Tier	Count	Focus
1	27	Grammar - particles, connectors, conditionals
2	24	Actions - single-kanji verbs for code operations
3	38	Nouns, Descriptors & Patterns

Japanese Accounting (accounting)

129 sentences across 9 tiers. Money/accounting vocabulary for a construction-industry ERP - enough to read and discuss the sales/cost/closing flow with Japanese colleagues. All readings verified against UniDic.

Tier	Count	Focus
1	18	Business flow - order, delivery, acceptance, billing
2	11	Recognition - sales/cost recognition, bookkeeping
3	12	Journals & accounts - debit/credit, ledgers, entries
4	16	Parties, billing & collection - bill-to, reconciliation
5	16	Balances & adjustments - AR/AP, offset, WIP
6	14	Closing & periods - period lock, carry-over
7	17	Standards, tax & compliance - recognition standard, invoice system
8	13	Construction cost & inventory - WIP, cost allocation
9	12	Discussion phrases - talking through the flow

Japanese Home Ownership (fudosan)

180 sentences across 12 tiers. 不動産・住宅購入 - the Japanese you need to buy land and build on it: phoning agents, reading listings, surviving the zoning and tax vocabulary, and asking city hall the right question. Mostly polite and keigo, with a written register for the language that only ever appears on contracts and municipal forms. All readings verified against UniDic.

Tier	Count	Focus
1	15	Phone basics - first call to an agent
2	15	Phone advanced - zoning, road and utility questions
3	15	Property listings - reading the basic figures
4	15	Listings advanced - boundaries, title, buried obstacles
5	15	Regulation basics - city planning areas, building standards
6	15	Roads & regulation advanced - frontage, setback, forest law
7	15	Tax basics - acquisition and fixed asset tax
8	15	Tax advanced - reductions, filing, valuation
9	15	City hall basics - reaching the right desk
10	15	City hall advanced - asking for a ruling
11	15	Pitfalls basics - the costs nobody quotes you
12	15	Pitfalls advanced - unregistered buildings, common rights

Agentic Lab Priority Path (agentic-lab)

554 sentences across 12 tiers. Tiers 1-5 are 179 sentences re-used from it-vocab: a short runway into an all-Japanese engineering team, with recovery phrases first, meeting structure second, the plain form you actually hear third, and subject-matter vocabulary deliberately last. That selection and its rationale live in docs/PRIORITY-PATH-agentic-lab.md.

Tiers 6-12 are authored here rather than selected. Tiers 6-8 are a production set for a spoken self-introduction to a new team, graded by what you say when. Tiers 9-12 are a comprehension set of Sino-Japanese work vocabulary, graded by how often the word turns up in day-to-day team writing, commonest first.

Note GUIDs are namespaced, so this deck and it-vocab can be installed together and scheduled independently.

Tier	Count	Focus
1	14	The survival set - keeping a conversation alive after comprehension fails
2	44	Meeting structure - openings, progress, estimates, parking a topic
3	66	Plain form, huddles and Slack - the register the polite tiers skip
4	28	Opinions, blockers and analysis - contributing rather than following
5	27	Explaining and deciding - structural connectives
6	22	Self-introduction: the frame - the nouns and the fixed opening and closing formulas
7	22	Self-introduction: background and domain - what you built, and in what field
8	28	Self-introduction: reading the room - how you work, what you work on, what you commit to, and stand-up ritual
9	49	Team vocabulary: core - the words in the most messages
10	57	Team vocabulary: common
11	90	Team vocabulary: regular
12	107	Team vocabulary: long tail

Minihongo Production (minihongo-speak)

830 sentences across 21 tiers, built from the minihongo core set by scripts/import_minihongo_speak.py. The other decks test whether you understand Japanese; this one tests whether you can say it. It is the only deck with the Production card described under Card Types. Where two rows would share an identical English front, only the core-word paraphrase gets the Production and Vocabulary cards; the other row keeps Listening and Reading, and its surface is listed beside the paraphrase on the back. See decks/minihongo-speak/REFERENCE.md.

Tiers are minihongo's own can-do goals rather than word frequency, and the six work can-dos come first: standup, incident, system explanation, demo, task planning, requirement clarification. Each tier opens with the dialog lines for that can-do and continues with the vocabulary those lines are built from.

Comprehension wide, production narrow: what you are asked to produce is the minihongo paraphrase, assembled from 231 core words, and the real Japanese word sits beside it on the back for recognition only. 会議 is there to be understood; 多【おお】くの人【ひと】が来【き】て考【かんが】える事【こと】 is there to be said.

Note GUIDs are namespaced, so this deck installs and schedules alongside it-vocab and agentic-lab even where a sentence is shared.

Tier	Count	Can-do
1	23	朝会で進捗を共有できる / I can give a standup update
2	24	障害を報告できる / I can report an incident
3	31	システムの仕組みを説明できる / I can explain a system
4	13	技術デモができる / I can run a technical demo
5	14	タスクを計画できる / I can plan a task
6	26	要件を確認できる / I can clarify a requirement
7	57	自己紹介ができる / I can introduce myself
8	46	店で買い物ができる / I can buy things at a shop
9	89	レストランで注文できる / I can order at a restaurant
10	58	道を尋ねられる / I can ask for directions
11	65	切符を買って電車に乗れる / I can buy a ticket and take the train
12	54	医者に痛みを伝えられる / I can describe pain to a doctor
13	12	薬局で薬を買える / I can buy medicine at a pharmacy
14	8	緊急時に助けを求められる / I can ask for help in an emergency
15	6	電話で予約できる / I can make an appointment by phone
16	59	市役所で住所を届け出られる / I can register my address at city hall
17	7	聞き返すことができる / I can ask someone to repeat or slow down
18	65	近所の人に挨拶できる / I can greet my neighbors
19	106	天気の話ができる / I can make small talk about the weather
20	23	円の値段が分かって言える / I can understand and say prices in yen
21	44	時刻が言える / I can tell the time

Download

AnkiWeb - install IT Vocabulary directly from Anki app

GitHub Release - manual download

Project Structure

decks/
  it-vocab/           # IT Vocabulary deck (15 tiers, 1516 sentences)
    deck.toml         # Deck config (name, IDs, tier definitions)
    translations.py   # Cloze word translations
    tier{1-15}-vocabulary.csv
  it-kundoku/         # IT Kundoku deck (3 tiers, 89 sentences)
    deck.toml
    translations.py
    tier{1-3}-vocabulary.csv
  accounting/         # Japanese Accounting deck (9 tiers, 129 sentences)
    deck.toml
    translations.py
    pronunciation_overrides.py   # readings that legitimately differ from UniDic
    tier{1-9}-vocabulary.csv
  fudosan/            # Japanese Home Ownership deck (12 tiers, 180 sentences)
    deck.toml
    pronunciation_overrides.py   # compound readings UniDic splits and mis-reads
    tier{1-12}-vocabulary.csv
  agentic-lab/        # Agentic Lab Priority Path (12 tiers, 554 sentences)
    deck.toml         # guid_namespace set: tiers 1-5 are also in it-vocab
    pronunciation_overrides.py   # readings UniDic loses when it re-tags a token alone
    tier{1-12}-vocabulary.csv
  minihongo-speak/    # Minihongo Production (21 tiers, 830 sentences)
    deck.toml         # production_card set: the only deck with the fourth card
    REFERENCE.md      # which minihongo rows feed which can-do tier
    pronunciation_overrides.py   # minihongo readings UniDic answers out of context
    tier{1-21}-vocabulary.csv
scripts/
  lib/config.py       # DeckConfig dataclass + loader
  create_deck.py      # Create Anki .apkg files
  generate_audio.py   # Generate TTS audio
  generate_furigana.py      # Fill Pronunciation furigana from UniDic
  generate_pitch_accent.py  # Pitch accent from UniDic
  validate.py         # Validate CSVs and audio
  pronunciation.py    # TTS preprocessing (shared)
  add_key_meanings.py # Generate English meanings
  test_tts.py         # Quick TTS test
skills/
  it-kundoku/         # Claude Code skill for IT Kundoku mode
tests/

Card Types

Listening - Audio-only front. No text. Forces pure listening comprehension for meetings and standups.

Reading - Japanese text front, no audio. Trains reading for Slack, PRs, and documentation.

Vocabulary - Key word blanked in sentence with English hint. Tests one word at a time.

Production - English meaning and situation on the front, nothing else: no Japanese, no furigana, no audio. You say the sentence aloud, then turn the card over for the furigana, the audio, the pitch accent, and the real Japanese word beside the minihongo construction. Opt-in per deck via production_card = true; only minihongo-speak uses it.

Features

• Pitch accent coloring (green=high, red=low) on key vocabulary
• Register badges (polite/keigo) on card fronts
• Neural TTS audio (Microsoft Edge ja-JP-KeitaNeural)
• Dark mode support
• Multi-deck support with shared pipeline

Build From Source

Requires Python 3.13+ and uv.

git clone https://github.com/KakkoiDev/nihongo-it-anki.git
cd nihongo-it-anki
uv sync

# List available decks
uv run python scripts/create_deck.py --list-decks

# Generate audio for a deck
uv run python scripts/generate_audio.py --deck it-vocab --all

# Create deck
uv run python scripts/create_deck.py --deck it-vocab --combined

# Female voice
uv run python scripts/generate_audio.py --deck it-vocab --all --female
uv run python scripts/create_deck.py --deck it-vocab --combined --female

All scripts accept --deck <slug> (defaults to it-vocab).

Releasing

Each deck is released independently with its own version tag ({deck}/{version}):

# Preview what will be released
uv run python scripts/release.py --deck it-vocab --version v5.1 --dry-run

# Create releases
uv run python scripts/release.py --deck it-vocab --version v5.1 --title "Tiers 13-15 + audit fixes"
uv run python scripts/release.py --deck it-kundoku --version v1.0 --title "Initial release"

# Draft release (not published)
uv run python scripts/release.py --deck it-kundoku --version v1.0 --draft

# Release notes from file
uv run python scripts/release.py --deck it-vocab --version v5.1 --notes-file RELEASE.md

Assets attached: {slug}-complete.apkg + individual {slug}-tier{N}.apkg files.

Scripts

Script	Purpose
create_deck.py	Create Anki .apkg files
generate_audio.py	Generate TTS audio for sentences
generate_furigana.py	Fill the Pronunciation column with 漢字【よみ】 furigana from UniDic
generate_pitch_accent.py	Generate pitch accent data from UniDic
check_pronunciation.py	Verify furigana readings against the UniDic dictionary (standard pronunciation check)
validate.py	Validate CSVs and audio files (runs check_pronunciation in advisory mode)
pronunciation.py	Furigana extraction, English-to-katakana conversion, TTS prosodic fixes
add_key_meanings.py	Generate English meanings for key words
test_tts.py	Generate audio for one or more rows to test pronunciation changes
import_minihongo_speak.py	Rebuild the minihongo-speak CSVs from a read-only minihongo checkout
release.py	Create GitHub releases per deck
migrate_guids.py	Migrate existing Anki collections from random GUIDs to stable sentence-based GUIDs. One-time fix for users who imported pre-stable-GUID builds.
migrate_deck_names.py	Rename subdecks in collection.anki2 from Tier N to Tier 0N. Detects orphan subdecks and can delete them with --delete-orphans.

Documentation

Deeper docs live under docs/:

• docs/tts-audio-debugging.md - ASR-based audit methodology, Edge TTS bug taxonomy, fix workflow, false-positive filtering. Read this if you hear a card audio that sounds wrong.
• docs/MIGRATE-DECK-NAMES.md - DB-level Anki migration guide covering both migrate_deck_names.py (subdeck rename + orphan cleanup) and migrate_guids.py (GUID + model-id retarget). Read this if you're upgrading an existing Anki collection.
• docs/REPORT-DECK-MIGRATION.md - Postmortem of the 2026-04-21 GUID + model-id live recovery. Anki schema gotchas (zstd anki21b, unicase collation, orphan note types, AnkiWeb sync divergence).

Skills

IT Kundoku

Claude Code skill for compressed IT Japanese communication. Install:

ln -sf "$(pwd)/skills/it-kundoku" ~/.claude/skills/it-kundoku

Trigger with /it-kundoku. Uses single-kanji vocabulary with Japanese readings and minimal grammar for token-efficient technical discussion.

Adding a New Deck

See Adding a New Deck in AGENTS.md for the full procedure, including ID registration and the guid_namespace requirement for decks that re-use another deck's sentences.

Credits

• Edge TTS - Text-to-speech (Microsoft Edge neural voices)
• genanki - Anki deck generation

License

MIT