Pulls real, human-made word-by-word lyrics from the lyric databases — picked by the song's language: LRCLIB + Musixmatch for Spanish/French/English & other Western, NetEase + KuGou for Chinese, PetitLyrics for Japanese — and matches them to this recording by duration. This is how proper KTV catalogs are made — far better than AI auto-sync. Pick the closest duration match.

Romaji guessed from kanji gets lyric readings wrong (風 → “fuu” instead of kaze, 私 → “watakushi” instead of watashi). This pulls the publisher's own furigana and stores it as this song's reading, leaving only kana→romaji — which is exact. Run it once per Japanese song, after the lyrics are final.

Separates the vocals into a KTV instrumental, then time-aligns to the isolated vocals. Character-by-character times every glyph for a precise karaoke wipe. Character, line-anchored does the same but pins every line to the timestamps already in your Lyrics field — the lines can't drift, and alignment only has to solve the timing within each line; it needs line timestamps such as [00:12.06] (Apple's [012.142s] is accepted too). Line-by-line just times each line's start (lighter). The last two need the song's lyrics saved first. Runs in the background (~10–20 min) — you can close this window and come back.

🔁 Re-sync characters only skips the slow separation entirely: when a KTV instrumental already exists, the vocal is just the mix minus that instrumental, so it re-times the glyphs in about a minute and keeps your existing instrumental untouched. Use it when the audio is fine but the wipe is off. It falls back to a full separation if the instrumental turns out not to match the recording.