Phoneme guides

What each phoneme token of each language-group model sounds like, with real example words, the languages that use it most, and suggested conventions for where its boundaries go. They are written for people who check or correct alignments, for example in the web interface, which shows the same text when you hover a phoneme. Agree on the conventions before several people annotate the same data.

Guide	Model	Languages
Latin	mbfa-latin	Latin-script languages without tone: English, German, French, Spanish, Portuguese, Polish, Turkish, Indonesian, Swahili and 40 more
Cyrillic	mbfa-cyrillic	Russian, Ukrainian, Belarusian, Bulgarian, Macedonian, Serbian, Kazakh, Tatar and more
Chinese (CJK)	mbfa-cjk	Mandarin (simplified and traditional script), Cantonese
Japanese	mbfa-japanese	Japanese
Vietnamese	mbfa-vietnamese	Vietnamese (Northern, Central, Southern)
Thai	mbfa-thai	Thai
Abjad	mbfa-abjad	Arabic, Persian, Urdu, Kurdish, Sindhi, Uyghur, Syriac
Brahmic	mbfa-brahmic	Hindi, Odia, Tamil, Sanskrit
Other alphabets	mbfa-other-alphabetic	Greek, Armenian, Georgian, Korean, Amharic
Tones		The tone numbers of Mandarin, Cantonese, Vietnamese and Thai

Which language uses which model, and the full list: language support.

The guides are generated by tools/build_annotation_guides.py from the G2P's output on the FLEURS dev transcripts. To change one, edit the descriptions in that script and run it again, not the pages themselves.