This is the baseline Kaldi recipe for unaccompanied karaoke singing using DSing corpus.
The DSing corpus is derived from the Smule Sing! 300x30x2 karaoke corpus [1].
It possesses three training sets: DSing1 with 15 hours of GB accent, 
DSing3 with 45 hours of GB+US+AU accents, and DSing30 with mix of accents from 30 countries.  
It was prepare as a speech recognition corpus by Gerardo Roa Dabike.
The recipe is in s5/


[1] Smule Sing! 300x30x2 Dataset, “https://ccrma.stanford.edu/damp/” accessed September 2018.