Japanese Subtext Trainer 1.0.2 Update
Behind the Japanese 1.0.2 rebuilds the speech pipeline to fix detached ending sounds and missing consonants, then adds a denser PC shell, calendar check-ins, analysis-to-next-stage flow, and four-panel manga scenes.
# Japanese Subtext Trainer 1.0.2 Update
Behind the Japanese keeps its 250-stage data-driven course while rebuilding speech generation, desktop layout, and learning history for a more reliable and compact listening experience.
## Full speech-reading reset
- Sentences, answer choices, and phrase tokens now store reviewable kana readings before they reach the offline model, while the interface continues to display the original kanji.
- The generator separates Misaki pitch metadata, normalizes special consonants, and rejects unknown phonemes. This removes the detached ending sound and prevents kyou from losing its ky consonant and becoming ou.
- Pipeline v4 forces the static audio library to be regenerated. The browser still never loads TTS during training, and the local model remains stopped with no autostart after the batch.
## A denser PC training shell
- The desktop tool adopts the game-area shell pattern: Back to Site at top left, the tool name at top right, and save synchronization centered in the frame.
- Repeated viewport-height constraints were removed, and the scene, questions, and analysis were rearranged to eliminate large unused gaps.
- Analysis now leads directly to the next stage. The Resources action is Start, and non-input headings, buttons, and card labels no longer become accidentally selected.
## Calendar check-ins and four-panel scenes
- Learning history is now a monthly check-in calendar with current streak, longest streak, total days, and recent activity, synchronized through a dedicated daily-activity table after sign-in.
- Every stage receives an original black-and-white four-panel manga scene matched to its prompt, with consistent characters, line work, screentones, and responsive placement.
