1) Read about TAVERN in our ISMIR paper: J. Devaney, C. Arthur, N. Condit-Schultz, and K. Nisula. Theme and variation encodings with roman numerals (TAVERN): A new data set for symbolic music analysis. In Proceedings of the International Society of Music Information Retrieval conference , pages 728–34, 2015.
2) See a full listing of TAVERN here.
3) Download TAVERN at github.