Thanks for your interest in SDCH.
There's no publicly released dictionary builder at present. To create
a good dictionary for a training set, one can find all strings of
length at least kBlockSize bytes that appear in multiple documents in
the training set, then measure the effect on target size (after
applying SDCH and gzip) of including each of those strings in the
dictionary. Some long strings that appear in every file in the
training set may still not be included in the dictionary because gzip
compresses them effectively.