The tone model in cantojam/data/model.json and the character table in
cantojam/data/char_tones.json are derived statistics computed from:
Jason Lee, A Corpus of Cantonese Popular Music, 2000-2020 https://github.com/jasonleeubc/Cantopop-corpus Study: http://hdl.handle.net/2429/84374 Licensed CC BY 4.0
105 songs, five per year from 2000 to 2020, transcribed in Humdrum **kern
with aligned melodic, textual, and phonetic spines. Every number cantojam quotes
comes from those 37,877 sung syllables.
The corpus itself is not redistributed here. scripts/build_model.py
regenerates the derived files from your own checkout. The derived files are
redistributed under CC BY 4.0, matching the source. See DATA_LICENSE.md.
Thank you to Jason Lee for transcribing and openly licensing the corpus. This project would not exist without it.
bill-iu, Canto-0243 (ONE·搵·韻) https://github.com/bill-iu/Canto-0243 Licensed CC BY-NC-SA 4.0 with additional terms
Canto-0243 is an excellent offline 填詞 workbench: 0243 / 02493 / 394052 tone code search, rhyme and initial anchors, synonym and antonym lookup across 181,220 entries. It is the tool that made clear what was missing, which is any notion of melody.
No Canto-0243 code, data, or lexicon was copied, adapted, or vendored into this project. cantojam shares no files with it and imposes no NC or ShareAlike obligation on you. The two are complementary: Canto-0243 answers "which words are tone-compatible with this string", cantojam answers "what does this melody allow, and does my melody fit my words".
If you write Cantonese lyrics, use both.
cantojam/data/colloquial.json was hand-authored for this project and is covered
by the MIT license along with the code. The corpus is drawn from mainstream
radio Cantopop written largely in 書面語, so common 口語 characters (唔 嘅 哋
冇 嘢 咗) barely appear in it. That file fills the gap.