expand German pronunciation-override lexicon

[semidark]

German Pronunciation Overrides — Call for Contributions

We maintain a pronunciation override file at ./misaki/data/de_overrides.json that lets the German G2P pipeline bypass espeak-ng for words it mispronounces — brand names, English loanwords, and German words with irregular pronunciation.

Current categories

Category	Purpose	Examples
brand	Tech/product brand names	claude, pytorch, kokoro
en	English loanwords common in German tech/media	api, github, pipeline
de_foreign	German words with foreign-origin pronunciation	ekstase, synthese, oase
aliases	Alternative spellings that map to an existing key	moetandchandon → moetchandon

How to contribute

1. Pick a word that espeak-ng mispronounces in a German context.
2. Add it to the appropriate category in de_overrides.json with the correct IPA phonemes.
3. Key format: Keys are matched after normalization — lowercased, Unicode NFKD with combining marks stripped, + → plus, & → and, @ → at, non-alphanumerics removed. So the key for C++ would be cplusplus.
4. Priority: On lookup collisions, brand > en > de_foreign.
5. Submit a PR with your additions.

What we're looking for

• More brand names: AI/ML frameworks, cloud providers, hardware, popular software
• English loanwords: Tech terminology commonly used in German speech (e.g. streaming, backend, framework)
• German words that espeak-ng gets wrong — especially words with French, Greek, or Latin roots
• Phoneme accuracy: Please verify your IPA transcriptions reflect natural German pronunciation of these words, not their original-language pronunciation (unless that's how Germans actually say them)

Tips

• You can test espeak-ng's default output with:

  espeak-ng -v de -q --ipa "your word here"