Anki — Col 1:15-20 deck

Operational context for the deck and its TTS audio output. The deck powers a "corkboard with red yarn" study workflow (per Art of Teaching) — browse cards on phone, listen to the audio walkthrough hands-free.

What's here

cards.md — gitignored, regenerable. 86-card source-of-truth export from the Anki deck, organized into sections (vocab, cross-refs, voices, voice memos, etc.). Read top-to-bottom to surface duplicates. Regenerate with export_anki_to_project.py after deck edits.
README.md — this file.

Outside the repo

The pipeline lives at ~/Library/Application Support/AnkiProgramFiles/. Run scripts with that dir's venv: ~/Library/Application Support/AnkiProgramFiles/.venv/bin/python.

Script	Purpose
`sync.py`	Push deck to AnkiWeb. Reads hkey from `~/Library/Application Support/Anki2/prefs21.db`. Desktop app must be closed.
`export_cards_for_voice.py`	Write `../cards_for_voice.md` — a TTS-friendly walkthrough (Card N. front … back). Sorts by NID, not `due`. See "Stable order" below.
`export_anki_to_project.py`	Write `cards.md` here. Section-grouped, scannable.
`add_cards_batch2.py`	Template for adding cards in bulk (idempotent on Front text).
`fix_duplicates.py`	Template for editing existing cards in place.
`render_cards_audio.py`	TTS render all cards → chunks in `../.tts_chunks/`, then concat to `../cards_for_voice.mp3`. Resumable.
`stitch_append.py`	Render only missing chunks and append to the existing mp3 via ffmpeg.
`tts_speakable.py`	Pronunciation substitution layer (Col→Colossians, v.17→verse 17, BP→Bible Project, LXX→Septuagint, …). Only substitutes book abbreviations when followed by a chapter digit, so common words pass through.

Collection path: ~/Library/Application Support/Anki2/User 1/collection.anki2 Deck name: Col 1:15-20

Stable order — DO NOT change

export_cards_for_voice.py sorts cards by NID (creation order). It was previously sorting by cards()[0].due, but Anki's due-value assignment is unstable when new cards are added — the export order shifted, the position-keyed chunk cache misaligned, and four cards ended up duplicated in the audio file. NID-sort is load-bearing. Adding new cards always appends to the end of the audio; existing chunks stay valid.

If you ever change the sort key, you must nuke ../.tts_chunks/ and full-re-render. Quota cost is ~40k chars (full deck).

ElevenLabs

Key: ELEVENLABS_API_KEY in ~/.zshrc
Voice: iUqOXhMfiOIbBejNtfLR — Oxley - Rich and Deep (American grandpa, warm baritone, narrative/storytime). Hardcoded in render_cards_audio.py.
Model: eleven_multilingual_v2 (handles Greek transliteration well: πλήρωμα reads correctly)
Format: mp3_44100_128
Tier: starter, 90,000 chars/month
Quota check: curl -s -H "xi-api-key: $ELEVENLABS_API_KEY" https://api.elevenlabs.io/v1/user/subscription

Common operations

Add cards — copy add_cards_batch2.py, edit the CARDS list, run. Then sync.py, then export_anki_to_project.py to refresh cards.md.

Edit cards — copy fix_duplicates.py, key by Front text, run. Then sync + export.

Re-render audio after content changes:

Identify affected card numbers (look at NID-ordered cards.md / cards_for_voice.md).
Delete just those chunks from ../.tts_chunks/NNN.mp3.
Run render_cards_audio.py — it'll synthesize the deletes and concat fresh.

Listen on phone — pull deck on AnkiMobile (browse view works best, not study), OR open cards_for_voice.mp3 via the viewer at http://100.83.250.41:8889/sermons/col-1-15-20/cards_for_voice.mp3 (Tailscale-only, has HTTP Range so iOS audio scrub works).

Card schema

Anki note type: Basic (Front, Back). Tags carry the structure:

col-1-15-20 — present on every card
Verse: v15, v16, v17, v18, v19, v20
Chunk type: vocab, greek, cross-ref, quote, concept, structure
Source: bp, piper, bell, willard, staton, peppiatt, comer, tyler, ligertwood, voice-memo, bkc, hntc, nicnt, tntc, …
Theme: cross-as-throne, firstborn, firstborn-from-dead, peace-blood, fullness, image-of-god, reconciliation, head-body, series-slogan, …

Back text uses HTML (<b>, <i>, <br>) — Anki renders it; the exporters strip/convert it. Footer convention: Source: <path> (the exporters strip this line before TTS so it isn't read aloud).

Current state (2026-05-13)

86 cards, deck synced to AnkiWeb
cards_for_voice.mp3 — 51:11, Oxley, pronunciation throughout
Quota used this month: ~74k / 90k (after full re-render)
Last full render did the audio + pronunciation cleanup; chunks 001-086 = NID order