Vectorised heuristic for picking vowel intervals out of a segmental
TextGrid tier without knowing the language. A label counts as a vowel
when, after stripping length marks, stress marks, tone digits/letters,
superscripts, ties, spaces, and combining diacritics, every remaining
character is an IPA vowel letter — so long vowels (aː), nasalised
vowels (ã), and di-/triphthongs (ai, iau) all match, while
anything containing a consonant letter (ang, pa, n) does not.
Di- and triphthongs need no special handling: the test applies to every
character of the label, so a label matches when all of its letters are
vowels – ai, au, iau, uai, ɔi, aːi and ai̯ all pass.
Offglides written as consonant letters are accepted too, since many
traditions spell diphthongs that way: j, w and ɥ count inside a
nucleus that also contains a vowel letter, so aj, aw, ja, jaw
and waj match, while a bare j or w onset does not. A vowel mixed
with any other consonant (ang, an, pa) never matches.
Syllabic nasals count as vowel-equivalent nuclei: a label whose letters are
all nasals and which carries a syllabicity mark (m̩, n̩, ŋ̩) matches,
since these are tone-bearing units in e.g. Cantonese (唔, 五). The same
nasals without the mark (m, n, ŋ) do not.
Details
Recognised base letters: the IPA vowel letters (a e i o u y æ ɐ ɑ ɒ ə ɘ ɵ ɛ œ ɜ ɞ ɤ ɪ ɨ ɔ ø ʉ ʊ ʌ ɯ ʏ ɶ ɚ ɝ and the near-close central ᵻ ᵿ), their ASCII
upper-case counterparts A E I O U Y, @ (schwa in SAMPA-style labels),
and precomposed accented Latin vowels, including the pinyin tone-marked
forms (ã é ü ā á ǎ à ū ǔ ǖ ǘ ǚ ǜ ...) in either Unicode normalisation.
Labels in other schemes (whole pinyin finals such as ang, X-SAMPA
consonant-bearing rhymes) should be selected with explicit labels instead
— see filter_interval_rows().
See also
filter_interval_rows() for subsetting f0 frames by interval.