The perl training script provided in the Demo includes the synthesis procedure through HTS. You could first try using it.
You need to generate the labels to synthesis and store them in lab files under data/labels/gen.
Then select the 4 steps below in the config file (if you stick to single mixtures) and run Training.pl.
$MKUNG = 1; # making unseen models (GV)
$MKUN1 = 1; # making unseen models (1mix)
$PGEN1 = 1; # generating speech parameter sequences (1mix)
$WGEN1 = 1; # synthesizing waveforms (1mix)