Lyrics for Chinese, Japanese and Korean songs are rarely found, and karaoke does not fit those scripts #705

The problem

Lyrics for Chinese, Japanese and Korean songs are rarely found, and when they are, karaoke does not fit those scripts.

On real LRCLIB data, only 11 of 22 Korean lookups found lyrics. For anyone practising a song in those languages, the Lyrics tab mostly says there is nothing to show, for songs that are in fact on LRCLIB.

Why lookups miss

The song is there. It is filed under a different spelling of the same names.

• Artist in another script. LRCLIB files a song under the name the artist releases under, often native script (周杰倫, 아이유, 宇多田ヒカル), while the track may know the artist in Latin (Jay Chou, IU, Hikaru Utada). The reverse happens too.
• Title in another form. Traditional and simplified Chinese (紅豆 and 红豆), full-width and half-width characters.
• Title translated. IU's "좋은 날" is on LRCLIB as "Good Day".

A plain text comparison treats every one of these as a different song, and the strict "same song, same artist" check that keeps wrong lyrics out then rejects the right one.

Why karaoke does not fit

Karaoke fills words as they are sung, and the idea of a word does not carry over.

• A Korean word was filled as if it were a single syllable.
• A Chinese or Japanese line containing one space fell back to filling whole phrases at once.
• Korean lines wrapped in the middle of a word.

Constraints for a fix

• The rule that lyrics are only shown when they really are this artist's song stays strict. Another artist's song of the same name must never be accepted, in any script.
• Names may only match through real aliases from MusicBrainz or Wikidata. No romanisation guesses: two different artists can romanise to the same Latin name.
• Traditional-to-simplified conversion needs a data table, and that table's licence has to ship with the app.
• No new dependency. A lockfile change forces every desktop user into a full runtime re-download.

Being worked on in the feat/song-identification branch.