Mimikara:日語説到對應詞語時,英文字幕逐詞顯示
Mimikara 耳から - Display English sub-titles when they are spoken
Mimikara 將日文和英文字幕檔轉成 .ass 字幕,讓英文單詞在日語對應詞語説出時逐詞淡入。它以時間配對字幕提示,使用 UniDic 分詞,並讓 LLM 配對英日詞語,再按 mora 數或強制對齊為單詞定時;透過顏色和不透明度變化,令英文仍可由左至右自然閲讀。
I'v watched many days worth of anime over my life and learnt surprisingly little Japanese, despite always chosing original language audio. I was hoping to learn more Japanese, but incidentally, without effort and think it may be because as we read thee English subtitles left to right - the translation doesn't correspond to the audio. (I know translation doesn't mean one-to-one subtitle to spoken word, still, over time one would learn, I hope, corresponding Japanese words for their English translations).
For example, Luffy's famous line:
海賊王におれはなる!Kaizoku-ō ni ore wa naru!
which translates to:
I'm gonna be King of the Pirates!
I always thought "Kaizoku-ō" meant King (because it sounds like German for emperorer: "Kaiser"), but actually it means Pirate!
So I have just started a project: mimikara, which only shows the English subtitle at the exact time the audio is spoken
It works by turning a Japanese and an English subtitle file into .ass subtitles where each English word fades in at the moment its Japanese counterpart is spoken, by pairing the cues by time, tokenising the Japanese with UniDic, asking an LLM which English word maps to which Japanese word, timing those words by mora count or forced alignment, and rendering the line with colour and opacity changes only so the English still reads naturally left to right.
來源:DEV Community · dev.to
