HTS version 2.1 includes hidden semi-Markov model (HSMM) training/adaptation/synthesis, speech parameter generation algorithm considering global variance (GV), SMAPLR/CSMAPLR adaptation, and other minor new features.