Extract lyrics from MIDI JSON. Concatenates text from all tracks, removes spaces, and replaces <SP> markers with newlines. When Merge Repeated is ON, consecutive duplicate characters are collapsed into one.
When Resegment is ON, the original <SP> phrasing is ignored and the lyrics are re-segmented: digits are converted to Chinese number chars, spaces/newlines/punctuation are stripped, consecutive repeated characters are merged, then CT-Transformer re-adds punctuation and the result is emitted as one sentence per line (every punctuation-delimited fragment becomes its own line). This is useful for re-flowing lyrics into natural sentence boundaries. Note: when Resegment is ON, Merge Repeated has no extra effect — the merge step is already part of the re-segment pipeline.
By ahkimkoo·Created 2 months ago·Updated 2 months ago· 3