Pitch accent: bigger, more detailed custom dataset
- New custom pitch dataset: 495,732 Surfaces, 512,835 Branches, 570,409 Accents.
- Pitch selection now uses more detail, including surface, multiple readings, and POS.
- Homophones and nearby forms stay distinct. Ambiguous matches are rejected instead of wrongly choosing another word's pitch.
- If no pitch entry exists, Jiwami shows the reading only. No invented pitch pattern or fallback.
Jiwami launches
The first version of Jiwami is now available.