numpy
onnxruntime
quebra-frases
langcodes
# Floored at the release that speaks a lect's cardinals: `dialect_numbers` asks it for the
# Hijazi, Gulf and Egyptian number words by the lect's ISO 639-3 code, and an older parser
# answers `lang="acw"` with the literary words and no error, so the flag would do nothing
# and say so nowhere. It also reads the colloquial spellings of numbers, from 0.20.4a1:
# Arabic writes a contracted teen a dozen ways -- خمسطعش, خمستعشر, خمسطاعش -- and below that
# version the parser read ten of them, all of the -اشر shape, so
# `normalize_asr(spoken_numbers_to_digits)` left most of the teens in a real transcript as
# words. Measured on the human transcripts of the SADA corpus: 1,167 one-word teens in 190
# spellings, of which the older parser read 66.
ovos-number-parser>=0.21.0a1
# Floored at the release whose Arabic `nice_time` reads the hour as a feminine ordinal
# with the fractions of the hour and the period words: `normalize_for_tts` takes the
# words of a clock time written in digits from it. Below 0.31.5a1, twenty to the hour
# and noon are not read the way the cited sources give them.
ovos-date-parser>=0.31.5a1
# Transitive, floored here because the combination is otherwise installable and breaks
# on import rather than on resolve. ovos-config below 3.5.1a1 imports
# `ovos_utils.xdg_utils`, which ovos-utils 0.0.14a5 does not ship, so `arbtok.plugin`
# raises ModuleNotFoundError while `import arbtok` succeeds and the distribution reports
# a healthy version. A worker pool whose initializer imports the plugin then hangs
# instead of failing. Measured: `arbtok==0.0.0a81 ovos-utils==0.0.14a5` resolves without
# complaint and reproduces it; with these floors it resolves to the working pair instead.
ovos-utils>=0.15.1a1
ovos-config>=3.5.1a1
orthography2ipa>=7.95.1a1
huggingface_hub
