DubTitlerr glossaries
Per-show name/spelling glossaries used by DubTitlerr
to correct proper nouns in transcribed dub captions — character names, places, and
recurring terms that ASR routinely mangles (Dothamingo → Doflamingo,
spondum → Spandam).
What's in a glossary file
One JSON file per show, named <Show> (<Year>) {tvdb-<id>}.json to match the show's
TVDB identity (a bare <Show>.json for shows without one, e.g. One Pace.json).
Fields vary a little by show, but the ones that matter to a contributor are:
show— the display name used in prompts.wiki— the MediaWiki API endpoint DubTitlerr's glossary-verifier queries for canonical spellings.names— the proper-noun list the repair/name-correction stage guards against false positives with.hard_fixes— exact known-wrong → known-right pairs, applied unconditionally.phrases— multi-word terms treated as a unit rather than word-by-word.unanchored_repair— settrueonly for a show whose copies carry no Japanese-audio subtitles for the LLM repair stage to anchor on; see the note in DubTitlerr's own README before setting this.
Everything here is facts and short phrases — names, places, terms — mined from public wikis and fansub tracks and verified against those same wikis, not transcribed dialogue.
Using a glossary
DubTitlerr reads whatever is mounted at GLOSSARY_DIR, never a copy bundled in its
own source tree. Point GLOSSARY_DIR at a clone of this repository, and a
git pull on the host is what keeps it current:
git clone https://github.com/XenaRathon/DubTitlerr-glossaries.git /path/to/glossaries
docker run ... -v /path/to/glossaries:/config/glossaries -e GLOSSARY_DIR=/config/glossaries ...
Contributing
See CONTRIBUTING.md. Short version: open a PR against the show's JSON file (or add a new one), and describe where the correction came from.
There is no CI gate on pull requests here, deliberately — verification means querying the show's own wiki, and a flaky check on a contributor's first PR would turn people away faster than an occasional bad entry would. Changes are reviewed by hand.
License
Not yet decided for this repository specifically — treat contributions as intended for use with DubTitlerr (GPL-3.0) until a data-specific license is chosen.