DALI datasets split used to train models presented in the paper Multilingual lyrics-to-audio alignment (ISMIR 2020). - View it on GitHub