Japanese Deep Translate

Build a sentence-by-sentence Japanese study page that a Chinese-speaking learner can read AND listen to. For songs, the page syncs with the original audio.

Two entry modes

Mode A — song by name. User says "拆解 YOASOBI 群青" or similar. Use scripts/fetch_song.py to auto-fetch LRC lyrics + original MP3, which gives you a timestamped skeleton JSON. Then fill in the per-word breakdowns.

Mode B — raw Japanese text. User pastes any Japanese. Skip the fetcher; build the sentence list yourself. No original audio, no timestamps — just per-sentence tables + TTS.

Output format — non-negotiable

For each Japanese sentence, produce a 5-row table. One word per column. Particles and sentence-final particles get their own columns — that is the whole point of this format.